⭕ Flux LayerDiffuse Empty Conditioning

Flux doesn't use a negative prompt the way SD 1.5 does - with an LLM-based text encoder, an empty conditioning is often the more correct "negative" than any text you'd type. But when you're running LayerDiffuse on Flux inside this pack, the empty conditioning has to be built in exactly the shape Flux's attention expects, which is where the standard nodes can trip you up. FluxLayerDiffuseEmptyConditioning exists to build that correctly: it takes your CLIP and emits an empty conditioning tensor sized and formatted for the Flux pipeline, ready to plug into the KSampler's negative slot.

How it works

Three inputs, all simple:

• clip (CLIP, required) - the Flux CLIP, same one you use for positive encoding.
• batch_size (INT, default 1, max 4) - match this to your latent batch.
• sequence_length (INT, default 256, range 77–512) - the token sequence length of the empty conditioning. This is the number to keep in sync across the LayerDiffuse pipeline.

Output: empty_conditioning (CONDITIONING). Wire it to the KSampler's negative input. That's the whole node - it's a utility, not a feature, and it behaves like one.

Where it fits

The README's quick workflow runs the KSampler with this empty conditioning as the negative, paired with a CLIPTextEncode positive (DualCLIPLoader type="flux"). If you've been using a text negative prompt and getting weird results, this is the pack's intended substitute - for LayerDiffuse's latent-transparency offset to behave, the conditioning side needs to be the clean empty tensor rather than a prose sentence you'd write for SD.

Why the sequence length matters

That sequence_length default of 256 is the value the pack's FluxLayerDiffuseConditioningFix also defaults to (target_length 256). If you change one, change the other - they're two halves of the same contract. Get them out of sync and you're back to the tensor-size mismatch the fix node exists to cure. If you're not sure, leave both at 256 and you'll be on the pack's happy path.

Install and caveats

Same pack install as everything here - ComfyUI Manager → search "DenRakEiw Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/DenRakEiw/DenRakEiw_Nodes
cd DenRakEiw_Nodes && pip install -r requirements.txt

then restart. You'll also want TransparentVAE.pth and layerlora.safetensors plus a Flux checkpoint before the broader workflow means anything.

The one thing to keep in your back pocket: this whole Flux LayerDiffuse subsystem is a small author's port of a technique whose upstream tooling froze back in 2024–2025. When things misbehave, the README's troubleshooting is short and real - tensor errors point at the CLIP type and the Conditioning Fix node, the "everything is glass" look points at LoRA strength. If neither fits, you're in DIY territory: this is not a heavily-supported pack.