To compare two sigma schedules in ComfyUI, hold the model, latent, prompt, steps, denoise, and RandomNoise seed constant, and change only the scheduler option on BasicScheduler. Save each output with its settings recorded. This isolates the scheduler choice as a comparison variable; it does not prove that one schedule is generally better.
Set up a controlled comparison
ComfyUI’s BasicScheduler produces a SIGMAS output using the connected model and the selected scheduler, steps, and denoise inputs. Treat the output sequence as the result of those settings together, not as an effect of the node’s name alone. See the BasicScheduler documentation.
Use this sequence:
- Choose one workflow and record the model, prompt, latent input,
steps, anddenoise. Keep the same sampler and all other connected inputs in both runs. - Set a specific
noise_seedonRandomNoise, and check that its control-after-generate option will not advance the seed between runs. The documentation describes the same seed as producing the same noise pattern; it does not establish that this alone guarantees identical results across every workflow. See RandomNoise documentation. - Run the first condition and record its exact
scheduleroption and output filename. - Change only
BasicScheduler’sscheduleroption. Run again with the same recorded inputs, then record the second option and filename. - Compare the resulting images side by side. Describe what is visibly different, such as edge placement or texture, rather than assigning an overall quality rank based on this pair.
When checking connections, use the displayed type labels, such as SIGMAS, MODEL, and NOISE, as the decisive cue. Node colors can help locate connections, but are not a substitute for their actual type labels.
A hypothetical acceptance worksheet
The table is an original record template with a hypothetical example, not a ComfyUI test result or a promised outcome. Its example assumes both runs used the same workflow inputs except for scheduler.
| Record | Shared conditions | Scheduler and output | Hypothetical visible observation | Decision |
|---|---|---|---|---|
| A | Model M; latent L; prompt P; steps 24; denoise 1.0; noise_seed 314159 | normal; compare_A.png | Main object appears slightly left of center; fine texture is visible on the roof | Keep as the baseline record |
| B | Same M, L, P, steps, denoise, and seed | karras; compare_B.png | Main object appears slightly right of center; roof texture looks different at this view | Record the difference; do not infer a general winner |
Before interpreting the observations, confirm that both rows refer to the same inputs and that the filenames match the outputs being reviewed. If an input differs, the pair does not isolate the scheduler choice; correct the record and rerun the comparison.
Limits and next steps after a mismatch
A fixed seed controls the noise pattern described by the RandomNoise documentation, but it does not make a comparison controlled if the latent, model, prompt, steps, denoise, sampler, or other workflow inputs also change. Keep the generated files: they let you review this pair without treating a rerun as a replacement for the saved output.
If results differ and you cannot identify why, compare the recorded conditions and node connections first. If a setting or connection is uncertain, check its visible value and type label before drawing a conclusion. If the conditions match, report the visible difference for this pair only. The cited documentation does not establish that a particular schedule will improve a particular image, nor does it provide a universal quality ranking.
Three quick questions
Does the same seed guarantee identical images?
The documented behavior is that the same seed produces the same noise pattern. That is not a guarantee that every complete workflow or later rerun will produce an identical image.
Should I change `steps` to make the schedules comparable?
No. For this comparison, keep steps fixed so that only the scheduler option changes.
What if both outputs look the same?
Record that no visible difference was noticed in this pair at the viewing scale used. Do not conclude that the schedules always produce identical results.
Sources and next step
- BasicScheduler - ComfyUI Built-in Node Documentation
- RandomNoise - 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.