CLIP Text Encode SDXL Enhanced (Detail Isolation) Section
Building a detail-isolation chain
- prompt_sections_in
- prompt_sections_out
This node is a section - literally one chunk of a prompt, in node form. Chain a few of these together and feed the chain into the primary CLIPTextEncodeSDXLEnhancedDetailIsolation node, which turns the whole thing into SDXL conditioning with per-stream Cutoff isolation. On its own it produces nothing you can sample with; it's a building block, and it knows it.
How the chain works
Each section has an optional prompt_sections_in input and a prompt_sections_out output, both of type DETAIL_ISOLATION_SECTION_CHAIN - a list of section descriptors in chain order. Leave prompt_sections_in unconnected and this section is the start of the chain. Each node appends its descriptor to whatever it received, so wire them in sequence and the order is preserved.
What a section carries
global_text- your natural prompt chunk. Doubles as the Cutoff target text.enhanced_text- the expanded version the model actually encodes at this position (the region text). Empty or equal toglobal_textand the section becomes a passthrough chunk: tokens in the base prompt, no isolation region.global_text_weight/enhanced_text_weight(default 1, range -10..10).clip_l_strength/clip_g_strength(default 1, range -10..10) - per-stream control; 0 excludes this section from that CLIP stream entirely.- The A1111 embedding trio (
support_a1111_style_embedding_text,remove_text_for_unsupported_embeddings,filter_known_a1111_embedding_tags_not_installed_locally) - each section carries its own, so you can mix settings across the chain.
Output: prompt_sections_out, which feeds the next section's prompt_sections_in, and eventually the primary node.
Install
ComfyUI Manager → search "Unified Conditioning", or:
cd ComfyUI/custom_nodes
git clone https://github.com/RandyHaylor/unified-conditioning-combine-with-timestep-range.git unified-conditioning-merge
Restart ComfyUI; it's under unified-conditioning-merge. No models, no pip deps.
Troubleshooting
- If you see a validation error about
DETAIL_ISOLATION_SECTION_CHAIN, you've fed the chain into something that isn't the primary...DetailIsolationnode - the chain type only plugs into that node (and into other section nodes). - Chain order is chain order. Section A's
prompt_sections_outinto Section B'sprompt_sections_inmeans A comes before B. It's directional, and flipping it changes the prompt. - An empty
enhanced_textis a feature (passthrough), not a bug - but it's the most common reason a "missing" region shows up in the final prompt.
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| global_text | STRING | — | |
| enhanced_text | STRING | — | |
| global_text_weight | FLOAT | 1.00-10–10 | — |
| enhanced_text_weight | FLOAT | 1.00-10–10 | — |
| clip_l_strength | FLOAT | 1.00-10–10 | 0 = exclude this section from the CLIP-L stream entirely. |
| clip_g_strength | FLOAT | 1.00-10–10 | 0 = exclude this section from the CLIP-G stream entirely. |
| support_a1111_style_embedding_text | BOOLEAN | true | — |
| remove_text_for_unsupported_embeddings | BOOLEAN | true | — |
| filter_known_a1111_embedding_tags_not_installed_locally | BOOLEAN | true | List can be modified in custom node folder: known_a1111_embedding_names_to_filter_when_not_installed_locally.txt |
| prompt_sections_inopt | DETAIL_ISOLATION_SECTION_CHAIN | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| prompt_sections_out | DETAIL_ISOLATION_SECTION_CHAIN | — |