Product / Library
From prompt to delivery, nothing gets lost.
A downloads folder forgets. The library remembers: everything you have ever generated, searchable by what it is and how it was made - with the recipe to make it again.
Build your library freeOne library, every recipe.
What is the Archibal library? It is a searchable record of every AI asset you have generated, where each file keeps the prompt, model, seed, and workflow that made it - with lineage between takes, upscales, and masters matched automatically by content fingerprint.
Search that reaches into the work
Find anything by prompt fragment, model, project, or source. "The neon flamingo take from March, the one made with Seedance" is a query, not an archaeology project.
Every file keeps its recipe
Prompt, negative prompt, seed, sampler, references, checkpoint - recovered from embedded workflow metadata or recorded at capture. Re-run a ComfyUI step by dropping its saved workflow back onto the canvas.
Takes stack, the keeper stays
Re-rolls group into stacks - "6 takes, same prompt". Pick the keeper and sweep the rest; the record still shows honestly that 6 takes were made and 1 was kept.
Lineage without bookkeeping
Fingerprints (perceptual hash + ISCC) link generations to their upscales, composites, and masters automatically. Fingerprint matches and human declarations are labeled differently - distinct claims, distinct words.
The Inbox
Captures that arrive without a project land in an inbox with machine-suggested destinations - the machine suggests, you confirm. Nothing is ever auto-filed across client boundaries.
A production record per file
Open any asset and read its story: identity and evidence grade, the recipe, the models behind it, the stages and reviews it passed through, and every delivery it shipped in.
Frequently asked
What can I search by?
Prompt text, model, provider, project, capture source, evidence grade, delivery status, and whether a file carries a recoverable recipe. Every field the capture layer records is a way back to the work.
Can I actually re-run an old generation?
When the file carries an embedded ComfyUI workflow, yes - download the workflow JSON and drop it onto a ComfyUI canvas to reconstruct the full graph, seed and all. For other tools you get the recorded recipe: prompt, model, and settings to copy.
How does lineage matching work?
Every file is fingerprinted (perceptual hash and ISCC). New arrivals are matched against your existing work, so a keyframe, its upscale, and the final master link into one chain automatically - and the record honestly marks stages that were never captured.