ModelMergeFlux1_motorway_edition

Flux is a transformer, and its merge node reflects that: ModelMergeFlux1 exposes a ratio for every embedder and every one of the 57 transformer blocks in the model. This is its Motorway-ed clone. If you're blending Flux checkpoints, this is how you do it with precision instead of a single strength dial.

The layout, in plain terms

• img_in, time_in, guidance_in, vector_in, txt_in - the five input embedders: image latent, timestep, guidance vector, pooled text, and token text. These decide what the model attends to, so they have outsized influence on conditioning behavior.
• double_blocks.0–double_blocks.18 - 19 dual-stream blocks where image and text tokens interact. The early blocks build structure and semantic understanding; this is the "composition" side of the model.
• single_blocks.0–single_blocks.37 - 38 single-stream blocks, the bulk of the compute, where the model refines detail. This is the "texture and finish" side.
• final_layer - the output projection.

And the trap, same as every node in this family: all ratios default to 1.0, so an untouched node hands you model2. The workflow is: set a base blend across the board (this is where most people start), sample, then raise or lower the double blocks versus the single blocks to steer composition-versus-detail. The KB's merge essay has the right mental model - merges combine parent strengths at the cost of consistency, and per-block merging is how you spend that cost deliberately.

Merging Flux is where per-block control actually pays for itself, because Flux fine-tunes (dev vs schnell, community LoRA-converted checkpoints, etc.) tend to differ in where they changed the weights. If you can see which blocks carry a style, you can graft just those.

The inputs

• INPUT_model1_key (default model1) and INPUT_model2_key (default model2) - the two Flux models.
• The embedder and block ratios, each 0.0–1.0, step 0.01, default 1.0.
• OUTPUT_MODEL_key (default MODEL) - where the merged model lands.

The wrapper keeps every ratio as a widget, turns the models into key strings, and keyed-writes the result. Runtime behavior matches stock ModelMergeFlux1.

Install

One install for the whole pack. ComfyUI Manager, search "ComfyUI_agilly1989_motorway", or:

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart ComfyUI. No requirements.txt, no model downloads.

Troubleshooting

• Output identical to model2? All ratios default to 1.0 - set a base blend before fine-tuning.
• The guidance-related embedders (guidance_in) matter most for distilled/guidance-distilled Flux variants; blending them badly can break CFG behavior. Keep them near your base blend unless you know why you're moving them.
• Merge Flux with Flux only. Double/single block counts differ across transformer models, so these ratios won't map to other architectures.
• 'MotorwayClass' object has no attribute 'hash_' = a model key isn't in the Motorway; check the loaders ran and the key spelling matches (case-sensitive).
• Pack caveats: BETA, and the newest published version (1.1.7) disabled clone generation, so on a fresh install these _motorway_edition nodes may be missing from the menu. They're present in the 1.1.0–1.1.6 builds comfy.icu indexed.