To find out whether two chained LoRAs changed an image because of their order or their strengths, compare those variables separately. Keep the checkpoint, prompt, generation settings, and LoRA files fixed; first compare A→B with B→A at identical strengths, then change one node’s strength at a time. ComfyUI documents that Load LoRA nodes can be chained and that each node has model and CLIP strength controls. Its documentation does not promise that one order will be stronger or better.
Define a fair comparison
Save a copy of the current workflow and label the two LoRAs A and B. Record their exact filenames, the checkpoint, prompt and negative prompt, seed if set, sampler, scheduler, steps, CFG, dimensions, and any other setting that affects generation. Record both strength_model and strength_clip for each node. The node documentation lists both controls, so changing only the model strength while overlooking a different CLIP strength would not be a clean comparison.
Use one consistent output question, such as “Does the jacket keep its requested color without losing the character’s requested pose?” Decide in advance what counts as a pass, a failure, or an ambiguous result. This is your evaluation rule, not a ComfyUI feature.
Run the order comparison first
- Set the chain to A→B. Keep the selected LoRAs and every recorded setting fixed. Generate and save the output with its workflow and run details.
- Reorder the nodes to B→A. Because the cited documentation does not specify how MODEL and CLIP state is transferred when nodes are rearranged, inspect the actual connections and each node’s
lora_name,strength_model, andstrength_clipbefore running. Confirm that the intended LoRA and its recorded values are attached to the intended node, then generate and record this output. Treat this as a conditional comparison of the two observed workflows, not as a documentation-backed guarantee that only order changed. - Compare the results against the same criteria. Record visible differences, including failures, without assuming that a difference proves the order caused it if other inputs changed.
- Repeat both orders with the same seed when the workflow exposes a seed control. Treat the matching seed as an attempt to match one recorded condition, not as a guarantee that every source of variation is controlled. For each order, record the repeat number, seed, output filename, and whether the result agrees with the first run. If your workflow or setup does not let you hold randomness steady, record that limitation and use several outputs per condition. A single pair may not distinguish an order effect from generation variation.
The multiple-LoRA guide demonstrates chaining two Load LoRA nodes and adjusting strength_model in both. It is an example workflow, not evidence that reversing the chain has a particular effect.
Change strength one node at a time
After the order comparison, choose one order and hold it fixed. Change A’s strength_model while leaving A’s strength_clip, B’s two strengths, and all other settings unchanged. Generate and record. Restore the baseline, then test B’s strength_model the same way. If you want to investigate CLIP strength, run a separate set that changes only one node’s strength_clip at a time.
The node reference describes typical everyday values for each strength control as 0 to 1, while listing a wider input range. Treat that as documentation, not a recommendation that any value will suit your LoRAs. Record the actual values used and avoid changing both nodes together: doing so cannot identify which change mattered.
Read the comparison matrix
The figures below are hypothetical observations, not measured results or recommended settings. They illustrate how to record a controlled comparison.
| Trial | Chain and changed variable | Hypothetical observation | Interpretation |
|---|---|---|---|
| 1 | A→B, baseline strengths 0.7 and 0.6 | Pose passes; jacket color is unclear | Baseline for comparison |
| 2 | B→A, same recorded per-node strengths after connection and parameter checks | Color appears clearer; pose still passes in repeat 1, but repeat 2 is inconclusive | Hypothetical possible order-related difference; record each repeat rather than selecting the favorable output |
| 3 | A→B, A model strength 0.5 only | Pose passes; color remains unclear | Possible A-strength difference; compare with trial 1 |
| 4 | A→B, B model strength 0.8 only | Color appears clearer; pose fails | Trade-off to assess, not proof of a generally better value |
For each run, retain the output, workflow file, ComfyUI version, LoRA filenames, checkpoint, and recorded settings. Keep unsuccessful and ambiguous samples: discarding them can make a weak pattern look decisive. For a practical decision, keep an order or setting only if its relevant difference recurs and meets your stated criterion. If results vary between repeats, report the result as inconclusive rather than assigning a cause.
When the result is inconclusive
First check that the checkpoint and exact LoRA files did not change, that each node still receives the intended model and CLIP inputs, and that strengths remained attached to the intended LoRA when you reordered the chain. Then repeat a matched pair or run more trials while keeping the comparison narrow. If randomness cannot be controlled, say so in your notes and compare several outputs per condition rather than selecting one favorable image.
A chained workflow provides a way to apply multiple LoRAs and control strengths; the cited documentation does not specify a universal rule for predicting order effects, nor does it establish that a visual difference in one output generalizes to other prompts or seeds. In the built-in node reference, strength_model and strength_clip are separate inputs. This guide treats each node as a unit for the main order-versus-strength comparison; it does not diagnose interactions between those two branches.
Common questions
Should I change both LoRAs’ strengths to the same value?
Not for the first comparison unless that is your intended baseline. Preserve each node’s existing value so the order test does not also become a strength test.
Does a higher strength always make a LoRA more visible?
The node reference describes higher values as producing more pronounced model adjustment effects. It does not guarantee the same visible outcome for every LoRA, prompt, or image. Compare outputs against your specific criterion.
What if reversing the nodes produces no visible difference?
Record that outcome for the tested workflow and conditions. It does not establish that order never matters; it means this comparison did not show a difference you could identify under those conditions.
How should I judge inconsistent repeats?
Treat the condition as inconclusive when repeated runs under the same recorded setup do not give a consistent pass, failure, or visible difference. Keep every output, note the repeat number and seed, and report the proportion or pattern of passes and failures instead of selecting the most favorable image. This is an author-recommended comparison method, not an official ComfyUI measurement.
Sources and next step
- ComfyUI Multiple LoRAs: Chain Loaders for Blended Styles
- LoraLoader - ComfyUI Built-in Node Documentation
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.