Concat Condition & ELLA Embeds(Deprecated, CombineClip instead)

This node's display name tells you everything: "Concat Condition & ELLA Embeds (Deprecated, CombineClip instead)." If you're starting fresh, use Combine CLIP & ELLA Embeds and move on. But if you've inherited an old workflow with this node in it, here's what it is so you can decide whether to swap it.

In the source, ConcatConditionEllaEmbeds is literally registered as an alias of the CombineClipEllaEmbeds class - the same Python class, the same combine function. It takes a CONDITIONING and an ELLA_EMBEDS bag and returns a merged bag with the CLIP conditioning stashed under an ella_clip_embeds key. It isn't broken; the pack just renamed it and kept the old name registered so saved workflows don't hard-crash. Same inputs, same output:

• cond - a CONDITIONING from the CLIP side.
• embeds - an ELLA_EMBEDS bag.
• Output - a merged ELLA_EMBEDS bag.

Why deprecate a perfectly working node? Mostly naming. "Concat" is misleading - nothing gets concatenated; the CLIP content is added to a dict. "Combine CLIP & ELLA Embeds" says what it actually does, and it's the canonical name the maintainers want you searching for. Functionally, the two are interchangeable, so the only reason to replace an existing ConcatConditionEllaEmbeds with CombineClipEllaEmbeds in a working graph is hygiene - every node kept alive for backwards compatibility is one more thing that can break when the pack finally drops the old names.

It serves the same purpose as the rest of the CLIP-composition path: get CLIP-vocabulary content (LoRA trigger words) into an ELLA pipeline that otherwise only speaks FLAN-T5. For that job the modern options are Combine CLIP & ELLA Embeds or just ELLA Text Encode with a wired clip.

Install is the pack install - ComfyUI Manager or git clone https://github.com/TencentQQGYLab/ComfyUI-ELLA into custom_nodes, pip install -r requirements.txt - plus the ELLA model in models/ella and FLAN-T5 XL in models/ella_encoder.