Add alternative ways ('TE:' and 'EMB:') to load embeddings in prompt …

[silveroxides]

…since using 'embedding:' is kind of lengthy and cumbersome to use.

This makes "embedding:", "EMB:" and "TE:" all valid identifiers when wanting to load embeddings in prompt.

Here is example if #14744 has not been merged

example_14744_notmerged.json

Here is example if #14744 has been merged (Which might still require me to make minor conflict resolution though)

example_14744_merged.json

Here is the embedding used in both workflows

qwen3vl_4b_duckonbear.safetensors

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 274b940d-3f2e-4177-b749-5067b6c416fb

📥 Commits

Reviewing files that changed from the base of the PR and between 7d78232 and 7605b99.

📒 Files selected for processing (1)

• comfy/sd1_clip.py

🚧 Files skipped from review as they are similar to previous changes (1)

• comfy/sd1_clip.py

---

📝 Walkthrough

Walkthrough

comfy/sd1_clip.py now supports multiple embedding prefix identifiers instead of a single hardcoded prefix. SDTokenizer adds an embedding_identifiers property that includes the existing identifier plus EMB: and TE:. tokenize_with_weights now splits text with a regex built from those identifiers, identifies which prefix matched each fragment, and derives embedding_name from the matched prefix.

Changes

Area	Change
comfy/sd1_clip.py	Added embedding_identifiers; updated embedding-prefixed token splitting and name extraction

Sequence Diagram(s)

Not applicable.

Estimated code review effort: 2

Related PRs: None identified.

Suggested labels: tokenizer, enhancement

Suggested reviewers: comfyanonymous

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: adding alternate embedding prefixes for prompt loading.
Description check	✅ Passed	The description directly explains the same embedding-prefix change and includes relevant examples.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@comfy/sd1_clip.py`:
- Line 538: The new embedding prefixes in sd1_clip.py are too generic and can be
misread from normal prompt text. Update the prefix handling in the logic around
embedding_identifiers and the parser that checks for preceding whitespace so
that "EMB:" and "TE:" only match with a stronger boundary or other
disambiguation, and do not turn ordinary prompt fragments into embedding
lookups. Keep the existing "embedding:" behavior intact while making the
abbreviated identifiers collision-safe in the embedding resolution path.
- Around line 537-538: The embedding prefix override hook is broken because
tokenize_with_weights now relies on embedding_identifiers instead of the single
source-of-truth embedding_identifier. Update sd1_clip.py so
tokenize_with_weights in the SD1 clip tokenizer derives its prefix checks from
self.embedding_identifier first, or builds the identifier list from that
singular attribute, preserving subclass/custom tokenizer overrides while keeping
the existing EMB:/TE: aliases.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 70e3fba5-8a98-44b6-8d29-d07774c713f9

📥 Commits

Reviewing files that changed from the base of the PR and between 77917ed and 2256608.

📒 Files selected for processing (1)

• comfy/sd1_clip.py

[alexisrolland]

I am not super fan of this:

• Enforcing one way of doing things brings some simplicity.
• embedding: is longer to type but it has the benefit of making things explicit.
• It also matches the folder name which is a nice hint.
• If we were to support this, I think a better solution would be to modify the identifier directly self.embedding_identifier = "embedding:"

I would suggest to close it.