fix(openai): downscale oversized inline images - #62
Open
danielkov wants to merge 1 commit into
Open
Conversation
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Downscale oversized inline images before OpenAI subscription Responses request encoding. Images that exceed the serialized media-field limit are converted to bounded JPEGs, while compliant images, hosted images, and image metadata remain unchanged.
Motivation
Large screenshots submitted through ACP could fail the entire turn with
Responses media exceeds 1048576 bytesinstead of reaching the model.Impact
Oversized inline images sent to the OpenAI subscription provider now proceed at reduced quality or dimensions rather than failing. Processing is limited to image input and enforces source-size, decoded-memory, pixel-count, nesting, and cancellation bounds.
Technical details
Provider-boundary normalization
Normalization runs immediately before OpenAI request encoding, which covers ACP, TUI, and tool-result images without adding duplicate client-specific implementations.
Bounded image processing
CPU-heavy decoding and encoding runs on a blocking worker. JPEG quality is reduced first, then dimensions are progressively resized until the complete base64 data URL fits the existing Responses field limit.