Feat/image model switch - #26164

Hey! Your PR title Feat/image model switch doesn't follow conventional commit format.

Please update it to start with one of:

• feat: or feat(scope): new feature
• fix: or fix(scope): bug fix
• docs: or docs(scope): documentation changes
• chore: or chore(scope): maintenance tasks
• refactor: or refactor(scope): code refactoring
• test: or test(scope): adding or updating tests

Where scope is the package name (e.g., app, desktop, opencode).

See CONTRIBUTING.md for details.

The following comment was made by an LLM, it may be inaccurate:

I found a potential duplicate:

Related PR:

• PR feat(llm): auto-describe images via vision fallback when active model lacks vision support #24382: "feat(llm): auto-describe images via vision fallback when active model lacks vision support"
  • URL: feat(llm): auto-describe images via vision fallback when active model lacks vision support #24382
  • Why it's related: This PR implements the same core functionality - automatically handling images when the current model doesn't support vision by routing to a vision-capable model for analysis and converting the result to text. This appears to be either a duplicate or closely related feature addressing the same need.

The current PR #26164 appears to be an enhanced or refactored version of the work done in #24382, with added features like the command palette switch for image models and the /image-models slash command.

Hi @SOUMITRO-SAHA! This is a really useful feature - many of us are eagerly waiting for it. Two suggestions to help this PR move forward:

1. Fix the title to follow conventional commit format, e.g. eat: image model switch with automatic vision fallback. This will remove the
   eeds:title label and help get maintainer attention.
2. Consider adding a �ision_model alias alongside image_model in the config for consistency with the naming in earlier discussions (PR feat(llm): auto-describe images via vision fallback when active model lacks vision support #24382).

Let me know if you need help with anything!

This branch has been upgraded to V2. We're no longer taking PRs for V1. If you think this is still relevant, please port it over and open it for V2.

— from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊

Ported to V2 as requested: #62534

The V2 port keeps the core idea — images sent to a vision-less model are described by an image-capable model and the description replaces the image — but follows V2's architecture instead of V1's prompt-time part rewriting:

• It hooks the request-preparation path (SessionModelRequest) exactly where V2 already replaced unsupported images with an error placeholder, so every request kind (primary steps, compaction, titles) benefits and the transcript stays untouched.
• Descriptions are generated once per image and cached, instead of rewriting session parts.
• The describing model resolves from experimental.image_model in config, else automatically (preferring the session provider's own vision models, then a known cheap vision family — never a stranger's flagship by release order).
• Gated behind the image_fallback feature flag.

The TUI image-model picker from this PR isn't ported yet since V2 selects models server-side; happy to follow up with a picker if the core lands.