Source: raw/ChatGPT_Can_Make_Videos_Now._Here_s_the_Full_Workflow.md — “ChatGPT Can Make Videos Now. Here’s the Full Workflow” (youtube.com/watch?v=5Z7_7Ca5x-8, uploaded 2026-10-27)
Creator: HeyGen official channel (@heygen_official); presenter introduces himself on camera as “Michael”
URL: https://www.youtube.com/watch?v=5Z7_7Ca5x-8
Duration: 11:17
Platform: YouTube
Skill level: operator
A HeyGen-made walkthrough that builds a complete clinic welcome video without leaving ChatGPT, using the HeyGen plugin. Steps covered: picking a presenter, putting your own personal avatar in a new outfit, rendering a talking-head video, generating B-roll with HeyGen Video 1, stitching it under a voiceover, translating it into other languages in your own voice, and building a no-presenter explainer with Video Agent. The clinic is a demo: the presenter says he is not a doctor, the clinic is not real, and the workflow applies to any business. Its most useful parts are the failure notes, where ChatGPT misreported a video’s length and cut clips short.
Key Takeaways
Setup
- Install once, about a minute. In ChatGPT: Plugins → search “HeyGen” → click + → connect your HeyGen account. The free plan works. Every video made this way also lands in your HeyGen library.
- Invoke it from the composer’s + menu → HeyGen, or by typing @HeyGen. The presenter recommends toggling on the “work” tab in ChatGPT’s home section first.
- Gotcha: start a new chat after installing. If you install mid-chat, ChatGPT will not see the plugin.
- Pick an export folder in your prompt so assets land in one place.
Four presenter options
- HeyGen stock avatar. Describe the presenter (gender, age, ethnicity) and say which avatar model, which orientation, and to show it in the chat. Do the same for the voice: describe it (“woman, friendly”) or name it. If you do not specify an avatar, ChatGPT offers an in-chat picker. You can also call one by its avatar ID.
- Photo avatar from one image. The presenter described a character, ChatGPT generated three matching looks, and one more prompt turned a picture into a talking video. It was then combined with cinematic B-roll playing over the welcome voiceover. His verdict: fine for a talking head, but “a photo only knows what you look like, not exactly how you move.”
- Personal avatar (a digital version of you made from real footage). It moves, gestures and sounds like you, and he says it gives the best natural motion. Used for the rest of the demo.
- No presenter. Video Agent (bonus section below).
New outfit without losing your motion
- Two ways to make a new look: have ChatGPT edit a photo (white coat, badge, stethoscope), or ask HeyGen to generate a new look inside your personal avatar group using an existing look as reference. He says either works.
- The line that matters: “Add this as a new look on my existing personal avatar. Don’t create a new avatar.” Leave it out and you get a photo avatar again and lose the natural motion.
Scripts and length
- ChatGPT can draft the script. If you wrote it, say “word for word”, because agents “sometimes get creative with your script.”
- The script sets the length. Asking for 15 seconds does not produce 15 seconds.
- Do not hand it an exact duration to hit. It may speed up the voice to fit. Ask it to check the duration after rendering instead.
B-roll with HeyGen Video 1
- The plugin exposes HeyGen Video 1: describe a shot and get a clip of “5 to 15 seconds, up to 2K.”
- The demo generated four location shots (drone exterior, reception, waiting area, exam room) with the clinic photo attached as a reference so every shot looks like the same building, and “no people, just a space.”
- A separate “cinematic clip of me”: four shots with hard cuts from one prompt, which he calls the same format HeyGen uses in its own examples. The video does not say which model rendered this clip.
Stitching: watch the output, not the summary
- Failure seen: asked to join two clips, ChatGPT reported a 23-second result. The real file was 4 seconds: each clip played for 2 seconds and cut mid-sentence. Lesson one: watch what it made; don’t trust the summary.
- Fix: clips without a voiceover were being cut short, so give every shot narration. Start on camera, then let the voice carry over the B-roll “like a documentary.” This worked when requested in the same thread (about 20 seconds total).
- On review, one cinematic clip had been sped up to meet the timeline. He asked for a re-do “at a slower pace” and got a better result.
Translation in your own voice
- Translate the finished video (Spanish and German shown). Three behaviors he highlights:
- It shows both translated scripts before rendering, which is the point to bring in a bilingual reviewer.
- It asks which voice you mean rather than guessing.
- Consent: the caption reads “it doesn’t ask for consent before it recreates a voice and matches the lips,” but the next sentence is “You can confirm you have permission, and then it runs.” The intended meaning is almost certainly that it does ask. Treat this as a probable caption error, unconfirmed.
- The original stays untouched; translations are new versions, not replacements.
Video Agent (no presenter)
- One prompt in, finished video out: script, scenes, B-roll, captions and music.
- Demo: he pasted the clinic’s new-patient FAQ and asked for a 90-second video with no presenter. Before rendering, it showed a scene plan of eight scenes with every line of narration. He approved it in chat. Each scene can then be edited one at a time in the same thread.
Try It
- Install the HeyGen plugin in ChatGPT (Plugins → HeyGen → + → connect), then open a new chat.
- First render:
@HeyGen+ a stock avatar description + orientation + “show it in the chat” + a voice description + an export folder. - If you have a personal avatar, request a new look with the exact phrase “add this as a new look on my existing personal avatar; don’t create a new avatar.”
- Paste your script with “word for word”. Control length by editing the script, not by naming seconds.
- Generate 3–4 Video 1 B-roll shots with a location or product photo as reference and “no people.”
- Ask for the final cut with narration over every shot. Then watch it end to end and ask the agent to report the actual duration.
- Translate: review the translated script (or have a native speaker review it) before approving the render.
- For FAQ or explainer content with no presenter, paste the source text and ask Video Agent for a scene plan first. Approve or edit scene by scene.
Implementation
Tool/Service: HeyGen plugin for ChatGPT (calls HeyGen avatars, voices, HeyGen Video 1, video translation and Video Agent) Setup: ChatGPT → Plugins → search HeyGen → + → connect HeyGen account. Start a new chat. Trigger with @HeyGen or the composer + menu. Cost: The video says only that the HeyGen free plan works. Credit use per render, translation or Video Agent run is not stated. Integration notes: Outputs sync to the HeyGen library. Avatars can be referenced by ID. Agent-reported durations can be wrong, so verify every render. ^[inferred] Driving HeyGen from Claude over the HeyGen MCP connector follows the same pattern; the Avatar V practitioner notes record a similar gotcha (name the avatar version explicitly or the agent uses your default).
Open Questions
- “Plugin” vs “app” naming: the transcript says “plugin” and a “Plugins” section in ChatGPT; OpenAI’s current name for this surface was not checked.
- “Work tab” in ChatGPT: what the toggle is and whether the plugin needs it is not explained.
- Consent wording in the translation step (see above): probable caption error, not verified against the video.
- Which model rendered the four-shot “cinematic clip of me”: not stated. It could be Video 1 or an Avatar Shots / Seedance path.
- Limits: maximum Video Agent length, number of translation languages available in ChatGPT, and plan or credit costs are not stated.
- Compliance for real practices: the demo dresses an avatar as a clinician for a fictional clinic. ^[inferred] A real practice would need its own review before showing a non-clinician in a white coat; see the compliance line in Patient-Journey Avatar Videos.
Related
- HeyGen Video 1 — the B-roll model the plugin exposes
- Patient-Journey Avatar Videos — another HeyGen clinic demo built in Video Agent, with a compliance line
- HeyGen Avatar V — the avatar model; notes on agents driving HeyGen over MCP
- HeyGen Skills Bundle — the coding-agent route to the same HeyGen surface
- AI Video Generators Compared (HeyGen-made test) — Video Agent vs single-shot generators
- HeyGen Avatar Shots + Seedance 2.0 — multi-shot cinematic prompting with your avatar
- HeyGen Voice — the voice-clone step behind the plugin’s narration
- HeyGen Personal Avatar Playbook — building the personal avatar the workflow starts from
- HeyGen Video Translation — the translation step in depth