Stillpass: Generated images checked at full size

The group shot looked fine in the preview. At full size, two players share the same face, the hands are broken and garbled text is baked into the photo, and you notice after the post is scheduled. Thumbnails hide what full-size exports show. Stillpass gives your agent a brief, reference, generate and QA loop so every AI image is checked before it is used.

What it does

• Brief first sets use and size (feed 4:5 at 1080x1350, story 9:16, deck 16:9, print 1:1), where text will sit, and the house image rules.
• References uses one approved reference per face, white-balanced product photos and the brand swatch; the logo is never generated.
• Clause prompts writes one clause per concern with the framing first, so a failure is fixed by changing one clause.
• One at a time queues jobs, respects tool cooldowns and saves each export straight into the project folder.
• Full-size QA builds a face sheet at 200% and hard-fails repeated faces, age outside the band, broken hands, cropped heads and any text, logo or watermark.
• Prompt pack saves prompts/<series>.md with one prompt per ID, an engine syntax table and a status column.

How it works

1. Write the brief and read the image rules file.
2. Pick an approved tool and check credits before the run.
3. Prepare references and write the prompt.
4. Generate one image at a time, then QA the full-size export.
5. Fix by clause, edit (background removal, crop per ratio, local overlays), and save the pack.

Frameworks & tools

Canva generate-image, Gemini image models, Bloom, Higgsfield, Midjourney syntax in the pack, PIL or sharp for crops and face sheets.

Why this beats prompting it yourself

A plain prompt generates a batch and shows thumbnails. Stillpass checks each image at full size against its reference, writes every new rule back into the rules file, and leaves you prompts you can rerun.

Use cases

• Poster people such as hosts, players and crowds with distinct faces.
• Deck and landing photos with heads fully in frame.
• Product shots on a clean background.
• Screenshot watermarks before sharing outside.

Known limitations

Does not put text, prices or logos inside generated images, copy licensed characters or artists, or make video. Uses only tools you approved and asks before any tool or spend that is not approved yet.