Repository navigation
fix: restore Demo moderation and image creation - #485
Merged
nezhyborets merged 11 commits intoNov 10, 2026
Merged
Conversation
Moderation now uses omni-moderation-latest, and ImageStore constructs image requests with gpt-image-1 and size options supported by that model. The image view displays progress, request errors, and decoded base64 images while retaining URL support. Starting a request cancels the previous subscription and clears stale results and errors; cancellation ends progress without displaying an error. The diff adds tests for request parameters, moderation success and failure, image errors, retries, base64 results, and cancellation. Refs MacPaw#478
nezhyborets
requested changes
Nov 9, 2026
nezhyborets
left a comment
nezhyborets
left a comment
Collaborator
There was a problem hiding this comment.
Hi, thanks for the contribution! Please see the comments.
Also, could you add a Demo: entry under Unreleased → Fixed in CHANGELOG.md, like the other Demo fixes there?
nezhyborets
requested changes
Nov 9, 2026
Collaborator
The API no longer lists gpt-image-1 for new keys, so the release check's image creation case failed. Add Model.gpt_image_1_5 and use it in ImageStore.
ImageCreationView decoded each b64_json and built a UIImage on every body evaluation, including every keystroke in the prompt. ImageStore now decodes them off the main thread in its pipeline and publishes decodedImages, which the view reads.
The map closure in images(prompt:n:size:) inherited its main actor isolation and crashed the isolation check when it ran on the URL session queue. Use a static function instead.
Each behavior gets its own test, so a failure names the one that broke, and the handler that never answers says so.
The API reference has no description of the model to copy, and the doc comment was not taken from it.
ImageURLProtocol and ModerationURLProtocol were nearly identical, and so were the body capture and client setup around them. MockURL in its own file replaces both: it is configured through its respond property and exposes the last request body and the request count.
nezhyborets
reviewed
Nov 10, 2026
The store now decodes b64_json to a CGImage with ImageIO off the main thread and holds one GeneratedImage array, so it needs no UIKit branches.
Contributor
Author
|
Much appreciated @nezhyborets, with omni-moderation-latest and gpt-image-1 the Demo's moderation chat and image creation should work again. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Moderation Chat rejects messages with an invalid-model error. Creating an image silently fails, leaving the user without an image, progress feedback, or an explanation. Moderation requests use the rejected text-moderation-latest model, while image creation omits the required model parameter. Image failures are only printed, and the view only handles URL responses, so it cannot display base64 image results.
Moderation now uses omni-moderation-latest, and ImageStore constructs image requests with gpt-image-1 and size options supported by that model. The image view displays progress, request errors, and decoded base64 images while retaining URL support. Starting a request cancels the previous subscription and clears stale results and errors; cancellation ends progress without displaying an error. The diff adds tests for request parameters, moderation success and failure, image errors, retries, base64 results, and cancellation.
Ran
xcodebuild test -scheme DemoChatfor MiscStoreTests and ImageStoreTests on an iPhone 17 simulator (iOS 26.4): 3 tests (11 test cases) pass with this change.The moderation half has a runtime regression test: with the source changes reverted, MiscStoreTests compiles and fails because the request sends text-moderation-latest instead of omni-moderation-latest.
The image half is covered by ImageStoreTests, which call the new images(prompt:n:size:) entry point and read imagesQueryError. Those APIs do not exist on main, so these tests cannot run against the old code.
Refs #478
Why
The October 27 Demo release check (#478) found that Moderation Chat fails on every message and that image creation appears to do nothing. This covers those two items; the other observations in that report are left for follow-up.
Affected Areas
Demo app only:
MiscStore.swift,ImageStore.swiftandImageCreationView.swift, plus the newMiscStoreTests.swiftandImageStoreTests.swift. No library code changes.