ChatCUT in ChatGPT · AI Video Editing and Motion Graphics by Prompt

Quick Summary

• ChatCUT is a third-party AI video editor connected to the ChatGPT desktop app through a plugin. It is not a first-party OpenAI video editor, even though the workflow is directed from a ChatGPT or Codex conversation.
• Marcin AI begins with an unedited talking-head clip, asks the agent to remove mistakes and pauses, then uses the transcript-aware timeline to place motion graphics where the spoken content calls for them.
• The strongest production lesson is progressive approval. He tests the visual language on five seconds, refines glow and translucent card treatments, expands to twenty seconds, then commissions the remainder only after the design direction works.
• The editor transcribes the recording, cuts gaps, smooths audio joins, converts an incompatible ProRes source, creates motion graphics, captures supporting screen visuals, adds sound effects, reframes the speaker and applies a color treatment.
• The result is not generated by one magic prompt. It is a directed sequence of edit, review and refinement requests, with a real editable timeline available when manual correction is needed.
• ChatCUT offers a free tier with five credits. Its current monthly entry plan is $25 for 100 credits, while annual plans and larger tiers differ. Host-agent reasoning and ChatCUT-hosted generation can be billed through separate allowances.
• The video's $640 API-equivalent figure describes the displayed language-model token usage. It is not presented as an actual ChatCUT invoice and should not be used as the cost of the finished edit.

Credit and the important reality check

The embedded tutorial was created by Marcin AI and published on October 1, 2026. The demonstration, source footage and creative experiment are his. This article uses the supplied transcript, the public video description and ChatCUT's official documentation to turn the test into a repeatable production guide.

The video's headline calls this a new ChatGPT video editor. More precisely, ChatCUT is a third-party editing product with a plugin that lets a supported desktop agent operate a ChatCUT project. You write directions in ChatGPT or Codex, while the editable timeline appears in ChatCUT's browser panel and remains available for manual adjustment.

That distinction matters for pricing, privacy, troubleshooting and support. A ChatGPT subscription does not automatically include every ChatCUT generation. The two services can contribute different capabilities and consume different allowances in the same workflow.

What ChatCUT can actually control

According to ChatCUT's official plugin documentation, the agent can import and organize media, cut and arrange a timeline, inspect footage, transcribe speech, revise captions, create motion graphics, generate video, images, voiceover, music and sound effects, then export a finished video or an editable handoff.

The key advantage is shared context. The transcript tells the agent when a phrase is spoken, the media pool tells it which assets exist and the timeline exposes the current edit. A request such as 'show the ChatCUT logo when I introduce the plugin' can therefore be attached to the relevant moment instead of becoming a separate manual note for an editor.

The tool does not eliminate the editor. It moves many mechanical operations into a conversational control layer and keeps a conventional timeline underneath. That combination is more useful than an opaque one-click generator because the project can still be inspected and corrected.

Step 1 · Install the supported desktop plugin

Use the current ChatGPT desktop or Codex workflow, not a normal browser chat. ChatCUT's documentation says to sign in to the ChatCUT web app, open the avatar menu, choose Agent Plugin and copy the ChatGPT/Codex installation prompt. The plugin can also be found through the desktop plugin marketplace when it is listed there.

Authorize the connection only on the page opened by ChatCUT. Never paste cookies, passwords or access tokens into the conversation. After installation, start a completely new task. The available tool list is fixed when a task begins, so the installation conversation does not gain the newly installed tools retroactively.

In the new task, explicitly say that ChatCUT should perform the edit. ChatCUT's current verification command is 'codex mcp get chatcut'. If tools are missing after an update, reopen a fresh task before assuming the installation failed.

Step 2 · Prepare the source package before prompting

Marcin starts with a raw talking-head recording containing mistakes, pauses and repeated takes. A stronger production package also includes a short creative brief, delivery aspect ratio, target duration, logo files, font names, brand colors, pronunciation notes and any screen recordings that must appear.

He adds a motion-design reference from Dribbble to communicate transparency, typography, shadow and animation preferences. Dribbble's motion graphics search, linked in the sources below, is a useful place to collect references for transitions, type movement, interface animation, card treatments and visual pacing before writing the creative brief.

Use those references for general design principles, not for copying another creator's exact sequence, artwork or branded identity. Save the original link, name the specific qualities you are studying and require an original adaptation for your own message.

Keep assets in a clearly named project folder. The first request should state which clip is the primary recording, which files are logos or overlays, what can be generated and what must remain unchanged. A good folder reduces agent ambiguity and makes revisions easier to reproduce.

Step 3 · Finish the spoken edit before designing graphics

The first pass has one job: turn the raw performance into a clean spoken sequence. Marcin asks ChatCUT to remove mistakes, pauses, gaps and filler, preserve natural sentence boundaries and smooth the result into one polished video before any intensive motion work begins.

ChatCUT transcribes the recording, identifies sentence breaks, removes dead space, adjusts dialogue levels and softens audio joins. When the uploaded 4K phone file in ProRes was not immediately readable, the workflow converted it instead of stopping at the import error.

Review this cut for meaning, not only speed. Automated silence removal can erase intentional pauses, jokes, breaths or emphasis. Confirm that no good take was removed, every sentence is complete, audio joins are clean and the pace still sounds human before treating the transcript as the timing map for graphics.

Step 4 · Approve the visual language in five seconds

The most valuable technique in the tutorial is the five-second prototype. Marcin asks for several relevant logos and screen elements in the opening only, then checks whether the reference has been interpreted correctly before extending the treatment.

His first revision adds more glow, shadow and animation. The next establishes a liquid-glass treatment for cards: translucent surfaces, gentle background blur and no opaque full-frame graphic covering the speaker. This converts vague taste into reusable design rules.

A short sample saves credits and review time. It also exposes common failures early: the wrong logo, unreadable typography, cards covering the face, motion that fights the dialogue or an effect that looks attractive in a still but feels exhausting in motion.

Step 5 · Expand to a twenty-second production sample

Once the style is approved, Marcin commissions the first twenty seconds with dense motion graphics, animation and sound effects. He deliberately asks for more visual activity than a normal edit because the video is a capability demonstration.

The agent uses the transcript to synchronize logos, graphical callouts and scene treatments with specific words. It also creates stylized assets such as animated logos, glow treatments and a green-screen-like world replacement. The result is far more coherent because the design system was locked before this expensive pass.

For normal client or educational content, reduce the density. Reserve movement for a change in topic, a claim that needs proof, an interface step, a statistic or a punchline. Visual abundance is useful in a demo, but editorial restraint usually improves comprehension and perceived quality.

Step 6 · Reframe, grade and direct the ending

The finishing request adds a cinematic color treatment, centers the speaker, places a blurred background behind a cropped foreground subject and uses small punch-ins around edit points or emphasized lines. Marcin tests this treatment on the same twenty-second section before applying it more broadly.

His final direction is shot-specific. At the line introducing ChatGPT, the video should show the plugin interface and installation flow. At 'step by step', the viewer should see a moving timeline. At the ending, the speaker enters a stylized pixel world for a clear final surprise.

This is the difference between prompting and directing. A useful note identifies the spoken cue, the desired visual, its purpose and any constraint. 'Make it cinematic' is weak. 'At 00:16, show the plugin installation panel, enlarge the relevant control and keep the speaker visible on the right' is actionable.

Is it really crazy cheap? The pricing in plain language

ChatCUT's pricing page currently advertises a free plan with five credits, basic editing and exports, AI images and motion graphics, captions and music, avatars, voice previews and a 60-minute lifetime web export limit. The separate ChatGPT plugin page lists a monthly entry plan at $25 for 100 credits. Annual billing and larger tiers use different effective rates, so always check the live page before purchasing.

Not every operation uses ChatCUT credits. Official documentation says uploading, organizing files, manual timeline editing, transcription, standard caption editing and normal video, audio, subtitle or XML export do not consume credits, although plan limits can still apply. ChatCUT-hosted AI video, image, motion, music, sound effect and voice generation do consume credits at rates that vary by model and task.

When Codex or another local agent controls ChatCUT, the host agent's reasoning uses the host subscription or API allowance. ChatCUT credits are charged only when the workflow calls a ChatCUT-hosted generation service. If the host supplies its own image generation, that generation follows the host allowance instead. Read the confirmation card and completed credit history rather than assuming every action belongs to one bill.

At 08:06 in the video, Marcin shows 30 million cached input tokens, one million fresh input tokens and 119,000 output tokens, then says the API equivalent would be about $640. That is a model-usage comparison shown by the host interface, not evidence that he paid ChatCUT $640 for the edit. The transcript does not provide the actual ChatCUT credits consumed by the complete project.

How to control cost without weakening the edit

Write one complete brief with the goal, source media, output format, constraints and reference files. ChatCUT itself recommends using attachments or selection references instead of repeatedly describing the same clip or timeline region.

Separate cheap decisions from expensive generations. Approve the dialogue cut first, then the visual system on five seconds, then a twenty-second sample. Generate the whole treatment only after those gates pass. This is more reliable than requesting a fully finished video and revising every mistake across the entire runtime.

Reuse the same conversation while its project context remains useful, but begin a new conversation for unrelated work. Check the preflight estimate before generated media and the completed transaction in Credits History after the job. A failed or revised generation can change the real cost per accepted second.

Where this workflow fits, and where it does not

ChatCUT is well suited to talking-head cleanup, educational intros, creator videos, product explainers, social cutdowns and sequences where graphics need to follow spoken cues. It is especially attractive when an operator can describe the desired result clearly but does not want to build every animation manually.

Premiere Pro, After Effects, Resolve and specialist audio tools remain stronger when the job needs exact keyframe curves, complex compositing, color-managed finishing, broadcast delivery, frame-accurate collaboration or extensive manual control. The realistic comparison is not 'AI or editor.' It is which parts of the edit can be delegated without lowering the standard.

For professional delivery, inspect the entire final render at normal speed and frame by frame around transitions. Verify names, logos, colors, captions, visual claims, music rights, sound peaks, face framing, safe margins and export settings. Keep the editable project and source files, not only the final MP4.

The complete ChatCUT production checklist

Before the task: install or update the plugin, start a new desktop task, confirm the connection, create a project folder and gather the raw footage, brand assets, reference links and delivery specification.

Pass one: transcribe, remove mistakes and unwanted silence, choose the strongest takes, smooth dialogue joins and approve the story and pacing.

Pass two: define the type, palette, card treatment, shadows, transparency, animation rhythm, safe areas and visual exclusions. Generate a five-second sample and revise it until the design system is clear.

Pass three: build a twenty-second sample with representative graphics and sound. Review relevance, timing, legibility and density. Record the approved rules in the project.

Pass four: finish the remaining runtime with cue-specific directions. Add reframing, restrained punch-ins, color, screen captures, sound effects and a deliberate ending.

Delivery: watch the complete export, verify factual and visual details, check credits history, save the editable project and test the final file on the destination platform.

Verdict · The real breakthrough is the editable conversation

The impressive part of Marcin's demonstration is not that a model can place a logo on screen. It is that one conversation can inspect the transcript, manipulate a real timeline, create supporting graphics, revise a design language and preserve an editable project underneath.

The workflow is fastest when the human remains the director. Approve the cut, establish the visual grammar in a short sample, expand only after it works and give precise notes tied to spoken moments. That process can remove hours of mechanical labor without pretending that taste, rights and final responsibility have disappeared.

Try the free tier on a short, low-risk clip first. Measure the host usage, ChatCUT credits, elapsed time, number of accepted revisions and final review effort. That is the only honest way to decide whether it is cheap for your kind of video.