FormatForge

1. Download — gallery-dl or yt-dlp saves a file on this server.
2. PocketBase upload — one input_media row per file (fetched_files).
3. Image embedding (Gemini) — if GEMINI_API_KEY is set, the raster is sent to gemini-embedding-2-preview; the vector is written to input_media.embedding and model name to embedding_model.
4. Vision prompt (OpenRouter) — if OPENROUTER_API_KEY is set, the same file is described; text goes to metadata.image_recreation_prompt.
5. Prompt embedding + prompts row — after a successful vision prompt, if GEMINI_API_KEY is set, the prompt text is embedded and a prompts record is created (prompt_text, prompt_embedding, prompt_original_media); metadata.prompts_record_id is set on the input row.

PocketBase admin: open dashboard · public API base https://formatforgeplus.com