Treat input, middle, and out as three separate controls to investigate, not one overall strength slider. ComfyUI documents them as blending ratios for different model parts, but it does not say which visual feature each section will change for every architecture. Start by checking that your two models are suitable for this node, define the observable goal, and change one ratio at a time while holding the other two and the generation conditions steady.
A controlled decision process
- Confirm the models and node inputs. In
ModelMergeBlocks,model1is the base receiving patches frommodel2. The node documentation definesinput,middle, andoutas blending ratios for corresponding model sections. The official reference declares both model inputs asMODEL; that shared type alone does not establish architecture compatibility. Consult the documentation for your particular model pair before assuming these sections apply. - Write an observable target. Describe a feature you can judge in outputs, such as a particular rendering quality, alongside what must remain stable. Avoid assigning a feature to a block before observing it; the available documentation does not map named image qualities to specific blocks.
- Make a baseline. Choose two compatible models, record their identities and settings, and generate a reference with fixed prompts, seed where available, sampler, steps, and other relevant conditions. Pick a baseline ratio for each block. Record the values and keep two fixed whenever testing the third.
- Run the block rounds. First vary only
input, then return to baseline and vary onlymiddle, then repeat forout. Use a small, recorded set of comparison values that the node accepts. Keep all other generation conditions constant; if a setting cannot be held constant, record that limitation. - Compare against the target and the guardrails. For each output, note target-feature evidence, overall coherence, and unwanted changes. Keep a block change as a candidate only when the observed trade-off suits the brief. Then test a combination built from those candidates, still using the same generation conditions. This comparison method is an editorial recommendation, not a documented ComfyUI test protocol.
Hypothetical acceptance worksheet
The figures and observations below are hypothetical planning data, not test results or expected effects. Suppose the brief is to retain model 1's composition while exploring a texture associated with model 2. The values are illustrative ratios; do not assume they produce the listed observations.
| Round | input | middle | out | Hypothetical observation | Decision |
|---|---|---|---|---|---|
| Baseline | 0.20 | 0.20 | 0.20 | Composition stable; target texture not obvious | Reference only |
| Input only | 0.40 | 0.20 | 0.20 | Texture appears more often; composition also shifts | Reject for this brief: the composition guardrail fails |
| Middle only | 0.20 | 0.40 | 0.20 | No clear target change; coherence looks weaker | Do not carry forward |
| Out only | 0.20 | 0.20 | 0.40 | Target remains unclear; contrast changes | Do not carry forward |
None of the illustrated single-block candidates meets both the texture goal and the composition guardrail, so this worksheet does not justify advancing a combined setting. Reconsider the candidate values or model pair before a separate combination test. The recorded observations do not establish a universal visual meaning for a block.
Limits and what to do when it fails
The node page does not specify a universal ratio range, a formula for how ratios are applied, an architecture compatibility list, or a guaranteed relationship between a block and a named visual quality. Consult documentation for the specific model and node version, and do not transfer an observation from one model pair as a rule for another. The official node page identifies the inputs and output but does not define every workflow's compatibility requirements.
If outputs do not differ clearly, check that the merged model is actually the one used for generation and that only the intended ratio changed. Repeat the comparison with a prompt and conditions that make the target observable. If one block appears to improve the target but damages a required feature, record the trade-off instead of assuming a combined setting will cancel it. Test a combination only after single-block observations, and reject it if it does not meet the original guardrails.
Three quick questions
Does a higher ratio always mean more of model 2? The documentation calls each value a blending ratio for model 2's corresponding section, but does not provide a universal interpretation of every value. Use supported values for your workflow and compare outputs rather than assuming a linear or guaranteed visual response.
Should I change all three values together? Not for diagnosis. Change one section at a time first so you can associate an observed difference with the recorded controlled change without claiming exclusive causality. Combine candidates only after reviewing those rounds.
Can these blocks be assigned to a specific visual feature? The supplied node documentation does not provide that mapping. Treat any feature association as a hypothesis for a particular model pair and validate it with controlled generations.
Sources and scope
- ComfyUI ModelMergeBlocks node documentation defines the node inputs, their broad section roles, and the base/patch relationship.
- The same official node reference lists the resulting merged
MODELoutput; it does not promise architecture-independent visual effects.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.