I’ve been using Airtable as a prototype + implementation surface for visual-asset workflows. The working AP-ASK demo is now public.
The key value proposition is that Airtable is not merely organizing or tracking an AI image workflow. Rather, Airtable carries the workflow end to end, with image generation inside Airtable >>
brief >> shots >> generate candidates >> review >> accept
1/ Products + References
The base holds the product truth + the visual references directing the work.
Each reference states the role it plays, what it may decide + what it may not decide.
2/ Briefs + Shots
The brief + its shots assemble the production standard:
- what the image is for;
- what direction it follows;
- what may vary;
- what must remain true;
- where the accepted image will be used.
3/ Image generation inside Airtable
From a shot, the user clicks Generate new candidate image.
Airtable then:
- creates the candidate record;
- preserves the current standard, prompt + source inputs;
- checks that the inputs are complete;
- requests the image through the Airtable AI image field;
- returns the generated image to that same candidate record.
The user does not need to create an empty record, open it + find another Generate control.
One action initiates the candidate-generation workflow.
4/ Candidate review
The generated candidates accumulate under the shot and are reviewed against the standard that governed them.
Each candidate keeps a record of:
- the prompt used;
- the product + reference images used;
- the production standard in force when it was created;
- its generation status;
- its review state.
Candidates are numbered within each shot, rather than across the whole base.
The accepted result appears first, while the alternatives remain available for comparison.
5/ Recorded acceptance
A reviewer can accept a candidate for a named use.
That creates an accepted-image record preserving:
- which candidate was accepted;
- the approved use;
- the standard it was judged against;
- who accepted it;
- when the decision was made.
So the architecture does not stop at organizing prompts or tracking outputs. Rather, it carries structured creative intent through image generation, candidate review + recorded acceptance inside one self-contained Airtable system.
The public interface is read-only. The underlying base can also be inspected + copied, so builders can examine the tables, formulas, relationships, interfaces + automations behind it.
This remains a working reference implementation rather than a finished product. I’m continuing to develop it in public and extend it across additional commercial image-production workflows.
I’d be interested in feedback on 3 questions:
- Is the brief >> shots >> candidates >> acceptance relationship legible without studying the schema first?
- Does generating + reviewing the images inside the same base make the decision path clearer?
- Where would this pattern break or need to change for your own production workflow?
demo https://airtable.com/appAd5MxDJ7PnVBue/shrktCKpBa1RIVSqG
inspect + copy https://airtable.com/appAd5MxDJ7PnVBue/shrIkFkFoMOUMF9Mw
project https://asset-pipeline.A-S-K.studio/