Any Key Cafe · Production record · October 28, 2026
When AI Video Generation Fails: Our Documented Runway Experience
A fireside message, ten delivered clips, and an unresolved dispute over whether completed generations supplied a usable production.
The message goes forward
We commissioned a satirical fireside address for Any Key Cafe’s fundraiser: Grandpa Tin Foil, portrayed by our dragon character, speaking to the audience from an armchair. The message and its original audio were ready. The intended film was meant to carry that recording, including the complete invitation to support the project.
We approved a ten-shot Seedance 2.5 batch through Runway Agent: 1080p, 16:9, native audio enabled, draft mode off. The approved durations totalled 114 seconds. The original soundtrack runs 109.559 seconds; the difference reflects rounded clip lengths. The batch consumed 7,752 credits, leaving 22. According to our production ledger, $70 in new credit purchases had been made during this production. That cash figure is not a claim that 7,752 credits have an exact $70 exchange value.
Darren reports that this exhausted the available production budget. We attempted a local edit, preserving the original audio and covering omitted performance with a donation card. We ultimately judged the result unsuitable for the intended release. The homepage now carries the complete message with reactive light bars.
Listen to the fireside message →
What we asked for
The approved prompts called for a fixed camera, a consistent dragon and room, fixed exposure within each shot, stable props, and precise synchronization to the supplied audio. Shot 01 expressly requested direct eye contact with the camera. Every final prompt is reproduced with its clip below.
Important context: the prompts also explicitly asked for “calm dark negative space on the right third.” Darkness alone therefore does not prove a violation. Our complaint concerns how much detail is lost, the visual changes between shots, and whether the delivered sequence can serve its intended production purpose. The later request for roughly 50% more ambient visibility was a proposed correction, not an instruction we pretend was in the original batch.
What the evidence supports
- Omitted supplied dialogue: shot 09 begins at the GoFundMe sentence instead of the preceding invitation present in our supplied audio. The omission is also present in the original master, not just the compatibility copy.
- Changed timing: shot 09’s generated speech does not preserve the supplied recording’s timing. Replacing the audio locally restores the words but does not restore the missing matching mouth performance.
- Eyeline: our review of shot 01 found an off-lens opening and later drift despite its explicit direct-camera instruction.
- Very dark regions and continuity: sampled midpoint frames show substantial darkness in the right third, with differing brightness between shots. The independently generated takes begin and end in different poses. These are visible production concerns; their acceptability remains an editorial judgment.
What we corrected
Our first automated review incorrectly suggested that the glasses disappeared in shot 01. Frame-level review of the master showed that they remain visible. We also did not verify a true multi-frame freeze in the originals. Apparent holds in our local edit arose from retiming and our deliberate donation-card cover. We withdrew both allegations in the support correspondence and do not rely on them here.
Black regions are not proof of missing video frames. We have not established that half the frames or half the duration are missing. The documented visual concern is near-black image regions and inconsistent lighting. The independently verified missing content is the opening spoken invitation in shot 09.
How the files were checked
All ten original exports are 1920×1080, 24 fps, HEVC Main 10 files. Our retained comparison report found video similarity of 0.9906–0.9927 and decoded-audio correlation of 0.9989–0.9997 between originals and their compatibility previews. The previews are compressed copies of the same performances, not replacement takes.
The lighting figures below describe one midpoint frame per clip, not a whole-clip exposure score. They measure decoded image luma on a 0–255 scale in the right third. They are evidence of darkness, not a universal threshold for defective video. The data and contact sheet are downloadable.
Lighting measurements (JSON) · All ten sampled frames · File provenance and SHA-256 manifest
All ten delivered clips
The players use the retained H.264 compatibility copies, with their delivered audio and without our local retiming, replacement soundtrack, brightness correction, or cutaways. Each original HEVC master is separately available for download; some browsers cannot play that format. We include the complete batch, including footage whose individual shortcomings are limited, so readers can judge the sequence rather than a selected worst moment.
Delivered clip 00 · 8.064 seconds
Signature: mouth closed, tiny conducting flourishes
Requested: The introductory musical signature asked for a closed mouth and tiny conducting gestures.
Observed and verified: The sampled right third is extremely dark. This clip establishes the opening look; this record does not establish a separate missing-dialogue defect in this musical introduction.
Production impact: Potentially usable as an isolated opening. It does not remedy the later omissions or the sequence’s changing appearance.
Compare the exact supplied audio (7.800s)
Read the submitted prompt for shot 00
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Shot 00 of 10 is MUSIC ONLY: [Audio1] is a bassoon and pizzicato theme starting about 0.15s, a little chime at about 5.25s, decaying through about 7.5s, then a short handover. There is no speech. His mouth stays fully CLOSED for the entire shot with a soft pleased closed-mouth expression. Performance: he quietly conducts from his seat with two small mock-regal wrist flourishes timed to the pizzicato, each with a fluid preparation, a small crisp accent and a natural recovery, plus one pleased head bob. By the end both hands settle naturally and he looks at the camera. No singing, no added sound, no invented dialogue.
Delivered clip 01 · 14.080 seconds
Goal and Grandpa Tin Foil
Requested: The goal and Grandpa Tin Foil introduction explicitly requested direct eye contact and grave authority.
Observed and verified: At the opening (00:00) our master review found an off-lens gaze; later beats also drift. The midpoint has the darkest measured right third of the batch. The glasses remain visible; the earlier disappearance claim was wrong.
Production impact: The camera relationship undermines the personal address we intended. We judged this a substantive performance problem, although the character rendering itself has attractive detail.
Compare the exact supplied audio (13.115s)
Read the submitted prompt for shot 01
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Dry mock-regal, grave delivery. Shot 01 of 10, spoken content: the goal, the "dead serious" line, and the sentence about the words Grandpa Tin Foil. Performance: direct eye contact with the camera and grave authority, one tiny authoritative nod on "dead serious", and one expressive but restrained open palm, with a fluid preparation, a small crisp accent and a natural recovery. Hands settle between thoughts. After the last spoken word the mouth closes and he settles.
Delivered clip 02 · 9.042 seconds
Kardashev scale through Internet
Requested: The Kardashev-scale passage was to follow its supplied audio with the same fixed composition.
Observed and verified: The midpoint retains more room detail than shot 01, though the right side remains very dark. Its starting and ending pose do not create a matched continuous take with neighbouring clips.
Production impact: Individual footage may be salvageable with editing. We do not claim a separately verified missing sentence or freeze in this clip.
Compare the exact supplied audio (8.894s)
Read the submitted prompt for shot 02
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Dry, conversational mock-regal delivery. Shot 02 of 10, spoken content: "I also realize that, according to the Kardashev scale," then humanity going from arguing with strangers on the internet. Performance: one explanatory hand gesture on the Kardashev scale clause, then one small dismissive wrist wave on the arguing-on-the-internet line, each with a fluid preparation, a small crisp accent and a natural recovery. Hands settle in the natural pauses. After the last spoken word the mouth closes and he settles.
Delivered clip 03 · 7.072 seconds
Galaxy and municipal road construction
Requested: The galaxy and municipal-road-construction passage continued the same room and character.
Observed and verified: The midpoint is markedly darker on the right than shot 02. The visual change is visible in the contact sheet and at the sequence boundary.
Production impact: The abrupt change in exposure is a continuity concern for our intended address. We have not established missing video frames here.
Compare the exact supplied audio (6.500s)
Read the submitted prompt for shot 03
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Dry deadpan delivery. Shot 03 of 10, spoken content: "to harnessing the energy of an entire galaxy on a schedule normally reserved for municipal road construction." Performance: a measured expanding arc of both hands on the entire galaxy, with a fluid preparation and natural recovery, then the hands stop and he gives a deadpan glance at the camera for the road construction line. After the last spoken word the mouth closes and he holds still.
Delivered clip 04 · 12.064 seconds
Funding invitation and qualifications lead-in
Requested: The funding invitation and qualifications lead-in used the supplied audio slice.
Observed and verified: The sampled right third remains dark; the independently generated pose differs from adjacent takes.
Production impact: This clip is included for completeness. A cut might be usable in another edit; we are not claiming every second of the batch is unusable.
Compare the exact supplied audio (11.580s)
Read the submitted prompt for shot 04
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Dry mock-regal delivery. Shot 04 of 10, spoken content: "Nevertheless, here we are.", the invitation to fund this modest and entirely reasonable undertaking, the GoFundMe link line, and the introduction to his qualifications. Performance: a dignified reset on "Nevertheless, here we are.", then one inviting open palm on the funding invitation with a fluid preparation, a small crisp accent and a natural recovery, while his posture lifts into the qualifications introduction. Hands settle between thoughts. After the last spoken word the mouth closes and he settles.
Delivered clip 05 · 14.080 seconds
Titles and glowing staff
Requested: The titles and glowing-staff passage requested a subtle localized flourish without a global exposure change.
Observed and verified: The retained midpoint shows a dark right third. This page makes no unsupported claim about a specific prop teleportation or glow failure.
Production impact: The clip must be judged in sequence. Its presence does not restore the omitted closing invitation or provide continuous matching poses.
Compare the exact supplied audio (13.390s)
Read the submitted prompt for shot 05
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames, subtle local firelight and the staff-tip magic below move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position and is never touched, lifted or moved. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Increasingly hoity-toity mock-coronation delivery. Shot 05 of 10, spoken content: his list of titles: super theorist, wearer of the foil regalia, bearer of the Scepter of Improbability, regent of the Wild Wood, then "And before anyone asks, yes, it glows in the dark." Performance: two or three gestures in total, not one per title. A palm flourish on the foil regalia; on Scepter an open hand presents the staff beside the chair without touching it; an outward sweep on regent. The chin lifts slightly with the titles, and he keeps fluid preparation, a small crisp accent and natural recovery, with hands settling between beats. SPECIAL MAGIC, timed to the words "it glows in the dark" and the chime in the audio: warm gold light gathers at the existing brass tip of the staff, a small elegant spiral of three or four glittering motes rises and fades, and one delicate final twinkle lingers. Brief, lavish but localized: no flash, no global exposure change, no staff morph or movement, no particles across his face. He gives a tiny satisfied sideways glance, utterly unsurprised, then returns to camera. After the last spoken word the mouth closes and he settles.
Delivered clip 06 · 15.072 seconds
Not worthy; construction schedule; Six years maybe
Requested: The deadpan “not worthy” and “Six years, maybe” exchange allowed a brief sidelong look and return to camera.
Observed and verified: The submitted prompt deliberately permits a sidelong look here. It would be inaccurate to label every off-camera glance a prompt violation. The midpoint remains dark on the right.
Production impact: Dry comic staging is part of the intended result. Our sequence-level concerns do not make each permitted acting choice a defect.
Compare the exact supplied audio (14.150s)
Read the submitted prompt for shot 06
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Brisk deadpan delivery. Shot 06 of 10, spoken content: "No, you are not worthy.", the question about a detailed five-year construction schedule, "No.", "That would be ridiculous.", and "Six years, maybe." Performance: a crisp wrist flick on "No, you are not worthy" with one raised eyebrow and no cruel snarl, then the hands settle. One thoughtful question gesture on the schedule question, with a fluid preparation and natural recovery. Completely still and straight-faced on "No." and "That would be ridiculous." On "Six years, maybe." a tiny shrug and a sidelong look, then back to camera. After the last spoken word the mouth closes and he settles.
Delivered clip 07 · 10.080 seconds
Human beings, AI, automation, energy, science
Requested: The passage about humans, AI, automation, energy and science called for warmer delivery and measured gestures.
Observed and verified: The midpoint shows the dragon gesturing with a dark right-side region. Starting and ending poses belong to an independent take.
Production impact: Potentially usable in isolation. Matching the full speech to the neighbouring performances remains the editing problem.
Compare the exact supplied audio (9.640s)
Read the submitted prompt for shot 07
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Warmer, sincere curiosity. Shot 07 of 10, spoken content: "But somebody has to start asking what becomes possible when human beings, artificial intelligence, automation, energy, science." Performance: his dry manner softens into sincere curiosity, with two or three fluid, measured gestures at the clause pauses, each with a fluid preparation, a small crisp accent and a natural recovery. He is still between clauses; hands settle naturally. After the last spoken word the mouth closes and he settles.
Delivered clip 08 · 10.080 seconds
Determination and sincere curiosity
Requested: The determination passage requested an earnest look to camera on “That’s what I’m trying to find out.”
Observed and verified: This midpoint is among the darkest sampled regions, with 97.4% of right-third pixels below luma 16. This is a sampled measurement, not proof of a black frame or absent video.
Production impact: The amount of lost visible room detail contributed to our decision not to release this sequence as the finished film.
Compare the exact supplied audio (9.530s)
Read the submitted prompt for shot 08
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Sincere, earnest delivery. Shot 08 of 10, spoken content: "and a sufficiently unreasonable amount of determination are pointed in the same direction. That's what I'm trying to find out." Performance: on "pointed in the same direction" he gently gathers both hands together with a fluid preparation and a natural recovery, then the hands settle, then he gives an earnest look straight to camera on "That's what I'm trying to find out." After the last spoken word the mouth closes and he settles.
Delivered clip 09 · 15.042 seconds
Closing invitation, Kardashev Three, Welcome aboard
Requested: The closing audio begins “If that sounds worth a few bucks, or if you simply want to see what Grandpa Tin Foil does next…” before the GoFundMe sentence.
Observed and verified: At 00:00 the delivered master starts at the later GoFundMe line. Compare the supplied audio immediately below with the delivered clip. The complete opening invitation is absent and subsequent timing differs.
Production impact: This is the clearest content failure. A local audio replacement can recover the words, but cannot manufacture the matching mouth performance. We tried a donation-card cover before choosing the audio-first release.
Compare the exact supplied audio (14.960s)
Read the submitted prompt for shot 09
Locked tripod camera: fixed lens, framing, focus and exposure for the entire shot, one continuous take, no camera drift, zoom, parallax or scenery warping. [Image1] defines the exact composition and identity: one friendly cartoon dragon seated in the armchair by the warm stone fireplace, staff leaning beside the chair, warm amber firelight, calm dark negative space on the right third, which stays calm and empty. Fireplace masonry, walls, floor, chair and resting props keep their geometry, scale and screen position. Only the dragon's performance, localized flames and subtle local firelight move. Identical dragon face, proportions, colors, scales, mane, goatee, rust-brown vest and brass medallion; the round gold glasses stay securely on his nose. Stable eyes with natural blinks. Restrained speech jaw motion, teeth exactly as in the reference, never enlarged, no broad toothy grin. Stable seated contact with the chair, no body sliding, no floating. Exactly two arms with consistent hands and digits. Gestures stay inside the frame, clear of his face and props. The staff stays leaning in its original position. No generated titles, captions, logos, transitions or extra objects, no new characters. Lip-sync precisely to [Audio1], his voice only, with exact audio timing. No invented dialogue, no singing, no new voice, no echo. Shot 09 of 10, spoken content: the warm donation invitation, "the GoFundMe link is below", then "Next stop: Kardashev Three.", a short pause, then "Welcome aboard." Performance: one warm, welcoming open palm on the donation invitation with a fluid preparation, a small crisp accent and a natural recovery. "Next stop: Kardashev Three." is quietly triumphant with a small nod. He holds the natural pause in stillness with a calm look. On "Welcome aboard." a soft smile and one small nod. Speech ends at about 12.54 seconds into the clip. From 12.54 seconds to the end the mouth is fully CLOSED and silent with a soft closed-mouth smile, only gentle breathing and natural blinks; during the music after the speech he does not sing or move his mouth. Hands settle naturally between thoughts.
What happened when we asked for a remedy
Times below are Pacific Daylight Time on October 28, 2026. The account is based on the saved production notes and the actual support email thread reviewed for this page. We summarize correspondence to keep private account details out of the public record.
- About 00:36 — approved batch. The ten-shot generation was approved at 7,752 credits. All ten tasks completed and delivered files.
- 02:04 — escalation. The support chatbot escalated the request for human review. We requested restoration of the batch credits or one corrected batch at no additional charge.
- 03:04 onward — original-file evidence. We reported the lighting measurements, eyeline concern and omitted dialogue, supplied task identifiers, and clarified the intended continuity across the production.
- 05:27 — corrections. We explicitly withdrew the glasses-disappearance and original-frame-freeze allegations after further master review.
- 08:58 — Runway Support response. Support explained that generative models interpret prompts and references rather than reproduce them exactly, and that credits cover processing for completed generations. It declined a credit refund for a completed generation that did not match the customer’s expectations. It recommended prompting resources, cost checks, approval settings, Tool Mode, and community help.
- 09:08 onward — disagreement and further review request. We replied that omitted supplied dialogue and timing changes were not merely a matter of taste. We requested supervisor review. Darren subsequently reiterated that the available production budget had been consumed and that we would document the results publicly.
The response did not specifically discuss the missing opening invitation, the measured lighting, or the exact timing comparison. That does not establish whether support inspected the files. We can report what the response says and what it leaves unanswered; we cannot report the reviewer’s private actions or intentions.
Runway’s stated basis for the refusal is consistent with its published credit-refund policy, which distinguishes generation errors from completed but unsatisfactory output. We understand that distinction. Our disagreement is that completion of processing did not supply a usable execution of this particular audio-led project.
Our business assessment
For this production, we consider the combination of unusable portions and the refusal of a corrective credit remedy an unacceptable service outcome. We have chosen to move future work elsewhere. This is our assessment of a documented experience, not a claim that every Runway customer receives the same result or that the company deliberately caused the defects.
The outcome so far
The complete source audio survives. The homepage presents it with a simple reactive spectrum rather than the intended character film. No replacement generation was purchased for this release. At this record’s preparation, no credit restoration, replacement batch, or cash refund had been confirmed. Any later resolution belongs in the status notice at the top of this page.
Provenance and scope
Sources retained by Any Key Cafe include the final approved ten prompts, exact audio slices, original HEVC exports, compatibility copies, midpoint images, lighting measurements, master-comparison report, and support correspondence. Original files remain unchanged in the private exit archive. The public manifest identifies the files supplied here with SHA-256 hashes. It does not include private account links, email addresses, or authentication material.
Our later eleven-shot correction proposal split the closing invitation, added neutral splice handles, and requested brighter ambient visibility. Those additions are a proposed remedy informed by the first result, not evidence that those exact revised instructions appeared in the original prompts.
What we consider an acceptable AI-generated video
We understand that generative video can contain visual inconsistencies and other imperfections. We do not demand flawless results.
For evidence of that, watch our published announcement film for The Buried Sky. Darren reviewed it, accepted its remaining visible glitches, and approved it for publication because the finished film still communicated its message.
Our dispute here concerns something materially different: omissions and visual problems that, in our assessment, prevented the fireside footage from fulfilling its intended production purpose. We accept imperfections. We do not believe that accepting imperfections means abandoning a reasonable expectation of usable output.
Watch The Buried Sky — book announcement →We have linked the existing approved release, without modifying its remaining glitches. This is a practical comparison from our production history, not a controlled experiment. Examine both productions and reach your own conclusion.
