AI video · Case study
CATACOMBES: one filmmaker, twenty minutes, and the problem of shots that must cut together
Start with the film and the director's published character and location material. See which shots demanded rework and which accidental angle survived. The reported inputs describe effort, not profit.
Published · Reviewed
BitShovel illustration · not film footageViews change and measure reach only. The 242 hours, 3,229 generations and about $3,385 spent are the director's account, without invoices, platform payouts or net profit.
Check this resultRead the context and editorial assessment
A decade ago, Kévin Mendiboure wrote a Parisian underground sci-fi horror idea that he could not finance or get produced. He later completed CATACOMBES independently with generative tools and released the entire 19-minute-35-second film on his channel. The useful question is not whether one person can press a button to make a movie. He had to write filmable dialogue, define people, places and props, find shots in low-resolution tests and check each generation against an active edit. His interview makes several successes and failures concrete; shot-by-shot rejects, full prompts, invoices and income records are not public.
Verified material
Watch the whole film first: what do views establish?
The creator's channel hosts the 19-minute-35-second film; about 145,000 views was a platform count on the check date, not a paying-audience count.
CATACOMBES is published on the director's channel in collaboration with French Horror Movies. The sci-fi disaster story moves through Paris above and below ground, with dialogue, pursuit, underground spaces and an ending. The whole film lets readers judge recognizable characters, shot continuity and sound in narrative context. When checked on 24 September 2026, the player showed about 145,000 views; that number changes.
The original film and view counter establish a public work and attention, not completion rate, paid viewing or an exhibition-rights sale. The video description credits Kévin Mendiboure for writing and directing and lists visual, music and finishing tools. Those credits clarify the maker's role beside the tools, not the project's accounts.
Sources for this section: Kévin Mendiboure: complete original CATACOMBES film
Public account
How a decade-old idea became a producible short
The director rewrote the story and dialogue before using new image capabilities for sci-fi scenes he could not previously afford.
Mendiboure says the sci-fi settings made the concept too expensive about ten years ago and that no producer backed it. He began learning image and video tools in 2023; in spring 2026 he judged newer video output close enough to the live-action look he needed and rewrote the idea as a short. That is one filmmaker's project and technical judgment, not evidence that every shelved script is now cheap to produce.
He describes beginning with a story and distinct dialogue voices, then planning sets and shots. He says François and General Moreau began from photos of himself and his partner and that he had rights to use those likenesses. The interview does not publish a complete script or a rights ledger for performers and music. Anyone borrowing the process must start with material they may use.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure
Public account
When characters change, their reference sheets must change too
Multi-angle character sheets, location references, props and start frames supplied continuity across shots.
The interview shows front, profile, rear and close views of characters with costume details. The director says he updated the sheets after François, Kane and ARES-7 were injured or covered with alien material, so later shots had the right state to follow. Sets were not improvised anew for each shot: he prepared references for corridors, underground halls and combat spaces, along with recurring props such as weapons and a helicopter.
The interview pairs a Paris street reference photo the director took with generated settings. He considers the start frame fed into the video tool decisive for a credible shot: an unconvincing character in the starting image is hard to rescue in motion. The page offers selected examples, not all sheets, start frames or shot inputs. Its credited pictures are viewable there, not openly licensed for republication.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure
Public account
Test cheaply at low resolution before paying for higher-quality takes
The director reports testing shots in a lower-quality mode before several high-quality takes; an accidental high-angle close-up changed his plan.
His approach was to find the motion and prompt in Seedance 2's faster, lower-quality mode, then switch to high-quality generation and choose among multiple takes. It was a way to limit spending under his tools and pricing at the time, not a universal rate or a one-pass success formula. He sometimes left a shot brief open enough to discover new camera positions or staging from multi-shot output, then decided what to keep.
In an armoury scene where François talks about his brother, the plan called for a profile shot. A generation instead produced a high-angle extreme close-up. The director judged it more expressive, kept it and revised the shot list. Success here included recognizing a useful accident and reconnecting it to the story. The interview does not publish original side-by-side files for the planned profile and selected close-up, so this is not a controlled prompt comparison.
Documented process
A production sequence grounded in the film and interview
Follow the interview's examples back into the film; the complete project files are not open.
People and places
Inspect multi-angle character sheets, prop references and the street-photo-to-generated-setting example.
View the interview's examplesLow-resolution tests
The director says he tested prompts and movement in a cheaper mode before high-quality renders; individual original takes are not public.
Read the director's test methodRevise the shot list
An accidental high-angle close-up in the armoury confession survived and changed a planned profile angle.
Read the shot decisionReturn to the cut
Watch the complete film for whether position and emotion survive the edit, rather than judging one frame.
Watch the complete film
Evidence limitThe shared examples are mostly character, set and prop references, without same-shot failed and repaired footage, full prompts or spend ledger. BitShovel links to the original pages and film.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure · Kévin Mendiboure: complete original CATACOMBES film
Public account
The hard part was not only spectacle: dialogue reverse angles broke continuity
The director says generated reverse angles crossed the dialogue axis, requiring generation and editing together.
The director says quiet dialogue scenes were often hardest. Reverse angles in three-person conversations could break the 180-degree axis rule, leaving the audience unsure who stood where. A shot might look fine alone yet jump direction when cut against its neighbor. While waiting the eight to eighteen minutes he says a Seedance render took at the time, he edited accepted shots, checked positions and refined upcoming ones. Those timings describe his experience, not a service guarantee.
He says a roughly ten-second helicopter transition took almost ten hours. The transferable check is whether a shot can cut to its neighbors and serve its story task, followed by a changed angle, shot or edit if needed. The interview supplies neither every failed helicopter take nor a controlled comparison proving that one prompt phrase solved it.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure · Kévin Mendiboure: complete original CATACOMBES film
Public account
242 hours and 3,229 generations describe effort, not earnings
The director reports time, generation count and spend, without priced labor, invoices, revenue or profit.
In the interview and a pinned comment under the film, the director reports working alone for about 242 hours and making 3,229 image and video generations combined. Those are attempts, not 3,229 finished shots or a success rate. His comment puts project spending at about $3,385 and separately discusses how an annual subscription is counted. The figures help frame the cost of iteration, but no itemized invoices, subscription allocation, compute log or priced labor make this a full production-cost account.
The public film can serve as a portfolio piece for audiences or prospective partners, but there is no disclosed client order, licence fee, platform payout, exhibition settlement or profit for this film. A Netflix discussion mentioned in the interview concerns another project; it is not evidence that CATACOMBES was bought or distributed by Netflix. Views and media coverage do not replace receipts.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure · Director's pinned production-input comment under the film · Kévin Mendiboure: complete original CATACOMBES film
Our analysis
For a first attempt, test a dialogue sequence that actually cuts
A page of story, three dialogue shots and an edit check reveal cost risks better than attractive standalone frames.
Three dialogue shots expose story, space and character continuity at once: an establishing view fixes positions, and reverse angles carry speech and emotion. The BitShovel exercise at the end of this page turns that test into a shot list, retained candidates, rough cut and usable-seconds record. Those outputs say more about a possible longer film than one attractive keyframe.
This three-shot exercise is BitShovel's trial derived from the director's experience, not his published reproducible recipe. A portfolio or client proposal also needs a record of script, likeness, music and asset rights, delivery scope, revisions, labor and generation spend. Enquiries, signed work, cash received and profit must be tracked separately. His method can inform a trial, but it does not grant rights to reuse his footage, audio or characters, or promise similar views or income.
Sources for this section: Creative Bloq: production interview with Kévin Mendiboure · Kévin Mendiboure: complete original CATACOMBES film
A BitShovel exercise · not the maker’s tutorial
Try a three-shot dialogue that actually cuts
A long film can hide problems between shots. BitShovel designed this two-person, one-location trial from the director's account of broken reverse angles and editing during generation.
Before you begin
- Use an original or licensed one-page scene; clear likeness, voice, setting references and music.
- Limit the trial to two people, one location, three shots and a cap on spend and attempts before planning a feature.
Map the room and the turn
Write what A wants and how B responds. Mark their positions, eyelines and a dialogue axis, then list an establishing shot, A's line and B's reaction.
- Keep
- A one-page scene, blocking sketch and three-shot list with framing.
- Check
- Someone new to the story can say who speaks first and whose mind changes; positions and eyelines agree across shots.
Prepare character and set state
Make front/profile and wardrobe references for both people, note held props and emotional state, and keep one fixed location reference. Update state after an injury or prop change.
- Keep
- Two character-state sheets, one set reference and rights notes.
- Check
- Across the three shots, each person, outfit, prop and room direction remains recognizable.
Test three shots cheaply and retain rejects
Make short candidates at an affordable setting. For each shot record references, actual input, tool version, time, spend and reason for rejection, stopping at the preset limit.
- Keep
- Candidates and an attempt log for three shots, retaining at least one rejected take per shot.
- Check
- Check faces, clothes, eyelines and action. A beautiful standalone take that cannot cut to its neighbour is still unusable.
Build a rough cut while generating
Place accepted shots on a timeline at once, check A's line and B's reaction, and revise the shot list first if screen direction or emotion jumps.
- Keep
- A three-shot rough cut, in/out points and rework decisions.
- Check
- An unfamiliar viewer, without notes, can locate each person, identify the speaker and describe the turn.
Count usable seconds and decide whether to continue
Total attempts, generated seconds, seconds kept, labour time and cash spend per shot, then ask viewers where the cut breaks immersion.
- Keep
- A usable-footage and cost sheet, two or three concrete viewer comments and a revision list.
- Check
- Keep editability and viewer understanding separate from public reach and actual income; this trial usually tests only the first two.
Make one final choice: If identity or space will not hold across three shots, simplify the set or angle. Stop at the sample if rework exceeds budget; extend the story only when the cut is coherent, understood and affordable.
This is a BitShovel exercise based on the interview, not Mendiboure's original shot list, prompts or a recipe for his reported 242-hour production. His shot rejects, full ledger and film income are not public.
Sources and limits
Where the material comes from
- Kévin Mendiboure: complete original CATACOMBES filmDirector-published YouTube film and credits; published 22 May 2026Published Checked
The complete 19-minute-35-second film, observed view count and credits are inspectable; views do not establish payment or income.
- Director's pinned production-input comment under the filmFirst-party director self-reportChecked
The 242 hours, 3,229 image and video generations, about $3,385 project spend and subscription treatment are his statements, not invoices or audited costs.
- Creative Bloq: production interview with Kévin MendiboureFirst-person director interview and credited examples, published 9 June 2026Published Checked
The director recounts the old concept, references, low-resolution tests, accidental angle, dialogue continuity and editing; no full reject archive, prompts or commercial settlement.