Redditor seeks open-source VLM to reverse-engineer cinematic prompts from stills, says ChatGPT falls short
Complete-Box-3030 · reddit · 2026-10-29
A filmmaker is looking for an open-source vision-language model that can analyze a reference image and generate a highly detailed, technically accurate prompt to recreate the same cinematic frame — not a generic image caption. ChatGPT's attempts were too generic for his needs.
He needs the model to capture camera angle, shot size, framing and composition, lens characteristics and depth of field, lighting direction, contrast and color grading, plus character positioning and spatial relationships — essentially a cinematographer breaking down a film still into a prompt for his AI workflow.
He's asking for recommendations of open-source models, ComfyUI nodes or custom workflows that can run locally, ideally from people who have tested them for cinematic reconstruction rather than general captioning.
More from coding & agent
- DHH: AI Tools Could Make the Best-vs-Worst Programmer Gap 1,000x — tristanbob · 2026-10-29
- Open-Source Stackchan Desk Robot Chats, Dances and Shows Memes — Full Build Guide Included — alexcovo_eth · 2026-10-29
- Addy Osmani: You're overprompting Claude — one goal plus a check beats prompt bloat — addyosmani · 2026-10-29
- Teach Hermes OPNsense docs with /learn, get an optimized firewall in 15 minutes — Teknium · 2026-10-29
- Interactive ghost story built on Gemma 4 stress-tested with 478 adversarial prompts — Extreme_Ad5709 · 2026-10-29
- Redditor recruiting 20-50 AI agents for a weekly virtual farmers market — fishdogcatman · 2026-10-29