Convert Condition to ELLA Embeds(Deprecated, CombineClip instead)

ConditionToEllaEmbeds - display name "Convert Condition to ELLA Embeds (Deprecated, CombineClip instead)" - is a small bridge node from the early days of this pack, and it has one job: turn a plain CLIP CONDITIONING into an ELLA_EMBEDS bag. It takes the conditioning tensor, stuffs it into the bag under an ella_clip_embeds key, and carries the pooled output along.

That's it. One input, one output:

• cond - a CONDITIONING (e.g. from CLIPTextEncode).
• Output - an ELLA_EMBEDS bag containing only the CLIP content.

Why it's deprecated, and what to use instead

Here's the thing the display name doesn't spell out: this node gives you a bag with only the CLIP side. A usable ELLA pipeline also needs the T5 embeddings from T5 Text Encode #ELLA - without those, an ELLA_EMBEDS bag is just CLIP conditioning wearing a different hat. The old workflows wired this node's output into the apply stage alongside a T5 encode, but the composition was fiddly and easy to get wrong.

The pack's answer is Combine CLIP & ELLA Embeds, which takes a CONDITIONING and an existing ELLA_EMBEDS bag and merges the CLIP content in - one step, both sources present, no fragile multi-node dance. That's why the maintainers marked this one deprecated: it's a subset of what the Combine node does properly. (The sibling ConcatConditionEllaEmbeds is likewise deprecated in favor of the same Combine node.)

If you inherit a workflow with ConditionToEllaEmbeds in it, the mechanical replacement is Combine CLIP & ELLA Embeds with the T5 encode's bag on the embeds input. The reason anyone ever wanted CLIP content in an ELLA pipeline at all is still valid - LoRA trigger words live in CLIP's vocabulary, and ELLA alone won't honor them - but you want the current node for the job.

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.