Video Shotcraft Review 2026: Shot Lists From Prompts

What Video Shotcraft Does

Video Shotcraft is a prompt-to-shot-list tool for pre-production. The repository at github.com/Vincentwei1021/video-shotcraft describes it as a Claude Code and Codex skill for cinematic product videos built on Remotion, with 152 shot recipe cards plus templates. The GitHub page lists 11,920 stars. The website is vincentwei1021.github.io/video-shotcraft.

Input is a script, a brief, or a URL. Output is a shot list: numbered shots, camera notes, lens and movement suggestions, and timing. Where it sits in the pipeline matters. It is not a video generator. It does not render frames, cut clips, or export an MP4. It sits between the brief and the edit, the same slot a human shot lister or a director's breakdown occupies.

That distinction is the whole review. If you want a tool that turns a prompt into finished footage, this is the wrong category. If you want a tool that turns a prompt into a structured list a shooter or an editor can follow, this is the category, and the question becomes how faithful the list is to the brief.

The 152 recipe cards are the interesting part. A generic LLM asked for a shot list will produce something plausible but generic: wide, medium, close, cut. Recipe cards encode specific setups — product hero at 85mm, slow push in, 24fps, three-second hold. The repository frames these as reusable recipes, which suggests the output is constrained toward known-good cinematic patterns rather than freeform invention. Whether that constraint helps or hurts depends on the brief.

Remotion is the rendering layer the recipes target. Remotion is a React-based video framework, so shot recipes that reference it are aimed at programmatic video, not a traditional NLE timeline. That shapes the output: timing is expressed in frames and seconds, not timecode ranges in a Premiere project.

Hands-On: 5 Prompts, 5 Shot Lists

I have not run this tool. What follows is a structural read of what the documentation and repository describe, plus how that maps to the comparison the brief calls for. I am not going to paste outputs I did not generate.

The intended comparison is five prompts through Shotcraft and two rivals, outputs side by side. The honest version of that comparison, done by someone who has the tool installed, would look like this: same brief, three tools, count shots, count usable shots, count edit rounds before the list is shootable.

What the docs describe as the input surface is a script, a brief, or a URL. A URL input is the one worth flagging. If you paste a product page, the tool has to infer the product, the audience, and the visual angle from page copy. That is a different task from a written brief, and the fidelity will differ. A brief that already states the angle will produce a tighter list than a URL that leaves the angle implicit.

For the five-prompt structure, the prompts that stress the tool are the ones with ambiguity. A 30-second product teaser with a stated hero shot is easy. A 90-second brand film with no stated visual direction is hard, because the tool has to make creative choices the brief did not. A prompt with a hard constraint — no people on camera, no location changes, single room — tests whether the recipe cards can bend or whether they default to their pattern regardless.

Two rivals in this category are a generic LLM chat and a manual shot list. A generic LLM chat, driven by a model like claude-opus-4.6-fast or gpt-5.5-pro, can produce a shot list from a prompt with no tool at all. The pricing snapshot lists claude-opus-4.6-fast at $30/M input and $150/M output, and gpt-5.5-pro at $30/M input and $180/M output. A shot list is a few thousand tokens, so the per-list cost on either is small. That is the real competitive pressure on this tool: the generic chat is cheap and already in your workflow.

What Shotcraft adds over the generic chat is the recipe layer and the Remotion target. The recipe layer constrains output toward cinematic patterns. The Remotion target means the output is shaped for programmatic rendering. If you are building product videos in Remotion, that shaping is the value. If you are handing a list to a human shooter, the Remotion framing is overhead you have to translate.

Output Quality Rubric

Judge a shot list on four axes. Coverage: does the list cover the whole script, or does it drop the middle? Shot variety: are there distinct setups, or the same medium shot repeated? Lens and movement specificity: does it say 50mm slow dolly, or just close-up? Continuity: do the shots hold together as a sequence, or contradict each other on location, wardrobe, or time of day?

Coverage is where recipe-based tools usually win over freeform chat. A recipe library has a fixed set of patterns, so the tool tends to walk the script and assign a pattern per beat. The failure mode is over-assignment: a 20-second script gets 14 shots because the tool fills every beat. That is a list you have to cut, not a list you shoot.

Shot variety is where recipe libraries can lose. If the library is 152 cards but the tool defaults to the same 10 for product video, the variety is nominal. The repository lists 152 shot recipe cards; how many get used in practice is a question only someone running the tool can answer. I would want to see the distribution across the five prompts before trusting the variety claim.

Lens and movement specificity is the axis that separates a usable list from a sketch. A list that says close-up is a note. A list that says 85mm, f/2, slow push in, 3 seconds is a shot. The recipe cards, by name, suggest the latter. Whether the output preserves that specificity or flattens it to shot size is the thing to check.

Continuity is the hardest and the most likely to break. A tool that generates shots one at a time has no memory of shot 3 when it writes shot 11. If the tool holds the whole list in context, continuity holds. If it generates per-beat, it drifts. The docs describe a shot list as the output, which implies a single pass, which is the better case for continuity.

Scoring each prompt on these four axes, 1 to 5, gives a 20-point per-prompt score and a 100-point total. That is the rubric I would use. I am not going to fill in numbers I did not produce. The rubric is the deliverable here; the scores belong to whoever runs the five prompts.

vs Manual Shot Listing vs Generic LLM Chat

Three ways to get a shot list. Manual, generic chat, and this tool.

Manual shot listing is the baseline. Time-to-first-list is however long it takes you to read the script and write shots, which for a 60-second spot is maybe 30 to 60 minutes for someone experienced. Edit rounds are near zero because you wrote it. Export format is whatever you use — a doc, a spreadsheet, a notebook. The cost is your time, and the quality is capped by your experience.

Generic LLM chat is faster. Time-to-first-list is one prompt, maybe two if the first is weak. Edit rounds are usually one or two, because the first output is generic and you have to push it toward your visual direction. Export format is whatever you paste it into. The cost is the token spend, which at the snapshot rates above is cents per list. The quality is capped by the model and your prompting.

Shotcraft sits between. Time-to-first-list is one prompt plus the recipe layer doing work the generic chat does not. Edit rounds are the open question: if the recipes land, fewer rounds; if they default to a pattern that does not fit your brief, more rounds than the generic chat because you are fighting the library. Export formats are the practical differentiator. The brief asks about CSV, PDF, and Notion. The repository describes templates but I cannot confirm which export formats ship. If the tool exports to a format your editor or Notion board can ingest, that saves a manual transcription step the generic chat does not. If it only outputs text, the export advantage is zero.

The honest comparison is this: generic chat is cheaper and more flexible, manual is slower but exact, and Shotcraft's bet is that a curated recipe library beats both on first-pass quality for product video specifically. That bet is testable with the five-prompt structure. It is not testable from the docs.

One more angle. If you already build in Remotion, the output being Remotion-shaped is a real advantage over both alternatives. A generic chat does not know your render stack. A manual list does not either. If the tool emits timing and camera notes in a form that maps to Remotion props, that is a workflow win that has nothing to do with list quality.

Pricing + Free Tier

The tool block lists pricing starting at $0/mo. That is the only price figure I have. The pricing page is the source of truth for credits, per-project cost, team seats, and refund policy, and I am not going to invent numbers for any of them.

What $0/mo starting tells you is that there is a free entry point. What it does not tell you is what the free tier includes. Credits, per-project cost, seat limits, and refund terms are all on the official pricing page, and the brief's own data-source list names that page with a capture date for exactly this reason. Prices move. A review that hardcodes a credit count is wrong within a quarter.

The comparison that matters for pricing is against the generic chat. A shot list through claude-opus-4.6-fast or gpt-5.5-pro costs cents at the snapshot rates. If Shotcraft's paid tier costs more than a few dollars per project, the recipe layer has to earn the difference in fewer edit rounds. That is the pricing question, and it is answerable only against the actual tier numbers.

For team seats, the relevant question is whether the recipe library is shared. If a small ad shop has one seat and the recipes are per-account, the team value is limited. If recipes are shared across seats, the library compounds. The docs would say. I cannot.

On refund policy, the only safe statement is that it is listed on the pricing page. Do not buy a tier expecting a refund you have not read the terms for.

Limits

Long scripts are the first limit. A recipe-based tool walks a script beat by beat. A 10-minute script has many beats, and the output list grows past what anyone shoots. The tool may handle it; the docs do not say where the ceiling is. If you have long-form content, test the ceiling before committing.

Non-English briefs are the second. The repository and site are in English, the recipe cards are named in English, and the model layer behind the tool is whatever you point it at. A non-English brief may work, but the recipe matching is likely English-keyed. If your briefs are not in English, that is a risk to check early.

No native NLE integration is the third. The tool targets Remotion, which is programmatic. If your edit happens in Premiere, Resolve, or Final Cut, there is no described bridge from the shot list into the timeline. You export text and rebuild. That is a real cost against a manual list written directly in your NLE's notes field.

The fourth limit is the category itself. This is pre-production. It does not generate video. Anyone arriving expecting a video generator will be disappointed, and the marketing around "cinematic product videos" can read that way. The repository is clear that it is a skill for Claude Code and Codex, which is a developer-facing framing, not a creator-facing one.

Fifth: it is a skill, not a standalone app. That means it runs inside Claude Code or Codex. If you do not use either, the tool is not available to you without adopting that environment. That is a bigger adoption cost than a web app, and it is the limit most likely to disqualify the tool for a solo creator who just wants a shot list.

Verdict

The tool is a pre-production shot list generator with a curated recipe library and a Remotion target. The repository shows 11,920 stars, which is real traction for a niche skill. The pricing starts at $0/mo, which lowers the cost of finding out whether the recipes fit your work.

I would not pick this for a large monorepo of video work where the edit happens in a traditional NLE and the briefs are in multiple languages. The Remotion framing and the English-keyed recipes are friction there. I would pick it for a solo creator or a small ad shop already building product videos in Remotion, where the recipe layer and the programmatic target both pay off.

The score is 3.5 out of 5. The category is right, the recipe library is the differentiator, and the pricing entry point is low. The deductions are the missing NLE bridge, the unconfirmed export formats, and the fact that a generic chat with a frontier model is a credible substitute for a fraction of the cost on short briefs.

How this review was researched

Sources: the vendor documentation and website at vincentwei1021.github.io/video-shotcraft, the repository at github.com/Vincentwei1021/video-shotcraft, the official pricing page, and the live AI model pricing snapshot for the comparison models. No hands-on testing was performed. The five-prompt structure and the four-axis rubric are proposed methodology, not reported results.

FAQ

Is Video Shotcraft a video generator?

No. The repository describes it as a Claude Code and Codex skill for cinematic product videos via Remotion. It produces shot lists and camera notes, not rendered video.

How much does Video Shotcraft cost?

Pricing starts at $0/mo per the tool listing. Credits, per-project cost, team seats, and refund policy are on the official pricing page.

Does Video Shotcraft export to CSV, PDF, or Notion?

The repository describes templates but does not confirm which export formats ship. Check the export-format docs before relying on a specific format.