Script to Veo video: turning a script into a batch of prompts

A 90-second YouTube Shorts script is roughly 12 cinematic shots. Producing them one at a time in Google Flow — paste prompt, wait 50-60 seconds, download, repeat — is half an afternoon of low-density manual work. Producing them in batch by chaining two extensions, ChatGPT AutoKuak to pull the shots out and Flow AutoKuak to generate them, takes 25-35 minutes and gives you your attention back. This is the real working pipeline, with the prompt template, a worked cost example and the mistakes that break it.

The operational problem for an AI creator

The typical loop for a faceless AI creator in 2026: write the script (by hand or with ChatGPT) → work out how it breaks into visual shots → paste each shot as a prompt into Veo → wait → download → edit. The middle steps — “break into shots” and “paste one by one” — are where the hours go. Producing 12 individual shots in Flow takes about two hours: 8-10 minutes per shot between writing the prompt and waiting for the render.

The bottleneck is not the model — it is switching between tools (Docs, ChatGPT, Flow), the mechanical repetition, and the loss of creative focus. Chaining two Chrome extensions that run those steps without you changes the economics: you stop operating tools and start deciding what gets published.

The pipeline in four steps

Step 1 — Script in Google Docs with a clear structure

The pipeline’s input is a script written in Google Docs (Notion exported to HTML works too, but Docs is simpler). Minimum structure:

• The video title as an H1.
• The hook (first 5 seconds) clearly marked.
• The body split into rough time sections (seconds 5-25, 25-50, 50-90).
• A CTA or close at the end.

The script does not need to describe the shots. That is step 2’s job. It only needs to be narratively clear and carry rough time markers.

Step 2 — ChatGPT AutoKuak pulls out the cinematic shots

This is where the operational leverage is. You set ChatGPT AutoKuak to page-to-ChatGPT mode with your Google Docs tab as the source. You use one prompt — not twelve — asking ChatGPT to analyse the script and return a structured list of shots. The prompt:

Analyse the script below and break it into 8-12 cinematic shots for Veo 3.1. Output JSON with this structure per shot: {"number": int, "duration_seconds": int, "veo_prompt": string using the SCAPI framework (Subject + Camera + Action + Plane + Illumination + Style), "edit_notes": string}. Each veo_prompt must be 100-150 words using real film vocabulary (close-up, dolly, golden hour, and so on). No generic filler. Script: [paste script here]

ChatGPT returns JSON with 8-12 objects. Each one holds a cinematic prompt ready for Veo. Copy it into a new Google Doc (or leave it in the chat — both work as the source for step 3).

This is where ChatGPT AutoKuak really pays off once you produce several videos: you set the prompt once and run the batch across five different scripts. ChatGPT extracts the shots from all five in sequence, with you touching nothing.

Step 3 — Flow AutoKuak generates the Veo clips in batch

The JSON ChatGPT returned already holds the 12 prompts. Move them (one per line, with a --- delimiter between each) into a new Google Doc or note. Open labs.google/flow with your project, set the model to Veo 3.1 Fast for iteration (or Veo 3.1 standard if you are going straight to final), and launch Flow AutoKuak in batch mode.

Recommended timer: 90-120 seconds between prompts for Veo Fast, 150-180 for Veo Quality. Clips download automatically to your Downloads folder as clip-01.mp4, clip-02.mp4, and so on.

Real time for this step: about 25-35 minutes for 12 Veo Fast clips. While it runs, you can be writing the next video’s script.

Step 4 — The editor assembles the final video

The 12 clips go into your editor (DaVinci Resolve, Premiere, CapCut). Because sequential order is guaranteed by the filenames, the first assembly is just dragging every clip onto the timeline. You add audio (stock music or AI voice from ElevenLabs), captions and soft transitions. For a 90s Short, editing runs 30-45 minutes.

Total end-to-end: 70-90 minutes for a Short, against 3-4 hours by hand.

---

A worked end-to-end example

Full numbers for a 90s Short about “how to wake your energy up in five minutes”:

Step	Time	Credit cost	Cost in $
1. Write the script (in ChatGPT, with prompts)	15-20 min	2-3 ChatGPT prompts	included in Plus
2. Pull out 12 shots with ChatGPT AutoKuak	5 min (1 batched prompt)	1 ChatGPT prompt (same Free allowance)	included
3. Generate 12 Veo 3 Fast clips with Flow AutoKuak (30 % discard)	25-30 min	~17 Fast clips = 340 credits	~$7
4. Edit in CapCut plus ElevenLabs audio	30-45 min	—	~$2 (stock music or ElevenLabs use)
TOTAL	75-100 min	~340 Veo credits plus AI allowance	~$9-10

Choose Veo 3 standard (50 credits per clip) instead of Fast and the cost rises to about $15-18 per Short. Go to Veo Quality and it is $25-30. The choice depends on what a video is worth: a tutorial Short for a 100k-subscriber channel justifies Quality; an experiment can iterate on Fast.

The key prompt template for step 2

The prompt you hand ChatGPT in step 2 is the heart of the pipeline. A fuller template with explicit parameters:

You are a film director who specialises in AI video. Analyse the script below and break it into cinematic shots to produce with Veo 3.1.

PARAMETERS:
- Number of shots: [8|10|12] depending on script length
- Visual style: [realistic cinematic | anime | comic | advertising | documentary | atmospheric sci-fi]
- Aspect: [9:16 vertical Shorts | 16:9 horizontal | 1:1 square]
- Duration per clip: [4 seconds | 8 seconds]
- Prompt language: English

OUTPUT FORMAT (strict JSON):
[
 {
 "number": 1,
 "duration_seconds": 8,
 "video_time": "0-8s",
 "veo_prompt": "[Shot type] of [specific subject with attributes], [action with verb + tempo], camera [movement type], lighting [description], style [reference]. Duration: 8 seconds. 9:16.",
 "edit_notes": "How it fits narratively and which transition to use into the next shot"
 }
]

KEY INSTRUCTIONS:
1. Each prompt must be 100-150 words (3-6 sentences).
2. Apply the SCAPI framework (Subject + Camera + Action + Plane + Illumination + Style).
3. No vague adjectives. Always be concrete.
4. Keep coherence between shots: same character, consistent palette, progressive mood.
5. Strong hook on shot 1 (the first 3 seconds of a Short decide retention).

VIDEO SCRIPT:
[paste your script here]

Save this as a prompt template in Notion or Google Drive and reuse it on every video. If you ship four videos a week, having this prompt nailed down compounds fast.

Typical pipeline mistakes and how to avoid them

• An ambiguous script: if the script is not narratively clear, ChatGPT will force the shots. Make sure the script has structure before asking for shots.
• ChatGPT returns invalid JSON: sometimes it wraps the JSON in commentary or markdown. Add to the prompt: “Return ONLY valid JSON, no text before or after, no markdown code blocks”.
• Veo prompts that are too abstract: if the prompts ChatGPT returns are generic (“a nice scene at dawn”), the problem is in the step-2 prompt, not in Veo. Ask for more concreteness.
• Skipping the pilot batch in Flow AutoKuak: without it you spend credits on a broken run. Always 2-3 prompts first, with the range set, before the full batch.
• Mixing timers between extensions: Flow AutoKuak needs long timers (90-180s); ChatGPT AutoKuak is fine at 15-30s. Set them per extension, do not reuse the same value.

When the pipeline is NOT worth it

Being straight about it, three cases:

• One-off long videos (10+ minutes): the pipeline is tuned for Shorts and reels (60-120s). Long videos with complex narrative are still more efficiently built shot by shot with manual iteration.
• Client commercial work with per-shot approvals: if every shot needs sign-off before moving on, a batch pipeline does not fit — approval cycles break the flow.
• When you are not using Veo: obvious, but if your video is 100 % stock plus text, the pipeline adds nothing. Flow AutoKuak is for creators generating AI visuals.

---

Pipeline variants for different kinds of creator

Variant: “faceless YouTube Shorts at scale”

Four videos a week, narrated with an AI voice. The pipeline as written, with Pro on both extensions for volume without friction. Monthly cost: about $300 for the full stack (see the Veo economics).

Variant: “narrative short film with a character”

One 60s video a week with a consistent character. Add a step 2.5: after pulling out the shots with ChatGPT, generate 8 reference images with Grok AutoKuak in Sequences mode (see Sequences mode). Those images feed Flow AutoKuak’s image-to-video. Result: a video with a coherent character instead of a random one.

Variant: “mixed tutorial, AI plus screencast”

An educational video mixing Veo clips (intros, transitions, hero shots) with real screen recording. The pipeline generates only the 4-6 AI pieces; the rest is recorded the traditional way. Cost drops to about $3-5 per video.

---

Frequently asked questions

Do I need Pro on both extensions for the pipeline to work?

No. Free gives 30 prompts per 24 h per tool. One Short uses about 1 ChatGPT prompt (extraction) and 12-17 in Flow (generation with discards). That fits inside Free for one video a day. Pro makes sense if you publish 3-4 videos a day.

Does it work without ChatGPT Plus?

Yes. ChatGPT Free lets you use the model (with OpenAI’s own rate limits). ChatGPT AutoKuak automates your account regardless of plan. Plus is recommended for smooth access without hard caps.

Can I swap ChatGPT for Claude or Gemini in step 2?

Operationally yes — other extensions automate Claude and Gemini in a similar way. The results differ: ChatGPT tends to be more structured on JSON output. Claude gives more creative prompts but is less predictable. Gemini is tuned for Google Workspace tasks. Start with ChatGPT and try alternatives if something specific bothers you.

What is the realistic time for step 3 if Flow is slow?

At peak hours (US/EU evening), Veo Fast can take 60-90s a clip. With a 120s timer in Flow AutoKuak, 12 clips take 24-30 minutes. Off-peak (early European morning), 25-30s a clip — the 12 take 8-10 minutes. Worth running big batches off-peak.

Does the pipeline work for horizontal 16:9 video?

Yes, identically. You change the “aspect” parameter in the step-2 prompt (to “16:9”) and the setting in Flow. Clips generate in 16:9 and assemble in a horizontal timeline just like 9:16. Useful for traditional long-form YouTube.

What do I do if the Veo clips have no audio?

Veo 3.1 standard generates native ambient audio (since May 2026). Veo 3.1 Fast produces silent video only. If your pipeline needs audio without going through the edit, set Veo 3.1 standard or Quality in Flow. If you are adding AI voice and music anyway, Fast (silent) is operationally quicker.

---

Start with one test Short

Recommended first exercise: write a 90-second script about something you know well (your hobby, your work, an idea you keep circling). Run the whole pipeline in one afternoon. Time each step. Wherever you got stuck is the step to optimise. To go deeper on the individual pieces: ChatGPT AutoKuak in batch covers step 2 in detail. Cinematic Veo 3.1 prompts goes deeper on the SCAPI framework. Batch generation in Google Flow with Veo is the Flow AutoKuak pillar. And if you want storyboards with a consistent character, Grok’s Sequences mode is the next layer.