This 3 Step Workflow Makes AI Filmmaking Easier Than Ever!
This workflow addresses facial detail degradation in AI video pipelines by introducing structured character sheets and dedicated LLM pre-production brains.
What this lesson covers
Dan Kieft breaks down a three-step pre-production and generation workflow for creating cinematic AI shorts inside Higgsfield and Claude. He demonstrates how to use custom character sheets and LLM director files to maintain character facial consistency and cut credit usage.
Key takeaways from the creator
AI-extracted notes, not independently verified product claims. Timestamp links let you check each point in the original video.
- 02:19 ↗
Uploading a structured director reference file into Claude or ChatGPT establishes an LLM brain for generating detailed scripts and prompts.
- 03:13 ↗
Analyzing cinematic stills on Shot Deck allows creators to extract camera lenses, lighting setups, and color grades for AI prompting.
- 05:23 ↗
Combining a close-up facial render with a headless body image in a character sheet prevents video models from degrading facial detail.
- 06:57 ↗
Using GPT Image 2 at 2K resolution delivers crisp, high-detail base facial portraits suitable for video generation.
Workflow outlined in the video
- Download or draft an MD reference file containing film directing principles to contextualize your LLM.
- Use reference platforms like Shot Deck to extract shot composition, lighting, and lens specifications for your prompts.
- Iterate on script beats and shot lists inside Claude before generating any media assets.
- Generate detailed front-facing facial portraits in Higgsfield using GPT Image 2.
- Assemble facial close-ups and body assets into a unified reference sheet to lock character consistency.
Before you use this workflow
These notes describe the source video at its publication date. Model access, pricing, connectors and interfaces may have changed. Check the original source and the provider’s current documentation before installing an add-on, connecting an account or spending credits.
ReelStack has not independently tested this workflow. Preview one representative shot and check motion, continuity and output quality before applying it to a full production. No result, cost saving or model capability is guaranteed.
Explore the tools
How to interpret the numbers
53,692 views captured 23 October 2026. ReelStack recommendation score: 23/100. Formula: radar-v5. This is ReelStack’s calculation, not a YouTube rating or a measure of factual accuracy.
Show the calculation
- 65% performance: 0.52× views vs 4 other channel videos observed at a similar age. View-evidence factor 98% (views / (views + 1,000))
- 25% freshness: 1/100 with a 14-day half-life
- Momentum pending: collecting daily snapshots; its 20% weight goes to observed performance, not free points
- 10% engagement: 8/100; likes + 4× comments, smoothed with a 500-view neutral prior