feat(flux): add scheduler selection for Flux models

[Pfannkuchensack]

Summary

Add support for alternative diffusers Flow Matching schedulers for Flux models:

• Euler (default) - 1st order, fast, current behavior
• Heun (2nd order) - Better quality, 2x slower (double model evaluations per step)
• LCM - Optimized for few-step generation (1-4 steps)

The scheduler can be selected in both the Linear UI (Generation Settings → Advanced) and the Workflow Editor (Flux Denoise node).

Backend changes:

• New invokeai/backend/flux/schedulers.py with scheduler type definitions and class mapping
• Modified denoise.py to accept optional diffusers scheduler, with automatic detection of sigmas parameter support
• Added scheduler InputField to flux_denoise invocation (version 4.1.0 → 4.2.0)
• Fixed LCM scheduler to use num_inference_steps instead of custom sigmas (LCM has its own sigma schedule)
• Fixed progress bar to show user-facing steps (1/8, 2/8...) instead of internal scheduler steps

Frontend changes:

• Added fluxScheduler to Redux state in paramsSlice
• Created ParamFluxScheduler component for Linear UI dropdown
• Integrated scheduler selection into buildFLUXGraph
• Added model-aware metadata recall for Flux scheduler

Related Issues / Discussions

QA Instructions

1. Select a Flux model (Dev or Schnell)
2. Open Generation Settings → Advanced Options
3. Verify the Scheduler dropdown appears with options: Euler, Heun (2nd order), LCM
4. Generate images with each scheduler:
   • Euler: Should produce identical results to previous behavior
   • Heun: Takes ~2x longer, may produce slightly different/improved results
   • LCM: Works best with 1-4 steps
5. Test in Workflow Editor: The Flux Denoise node should have a scheduler dropdown
6. Test metadata recall: Load an image with scheduler metadata, verify correct Flux scheduler is selected

Merge Plan

Standard merge, no special considerations.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable)
• ❗Changes to a redux slice have a corresponding migration
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

[Pfannkuchensack]

I did not test (the scheduler) changes yet. Frontend looks good.

[lstein]

I'm getting a validation error on the step progress callback for Euler and LCM. Heun is working ok. Something about the way steps are being counted, I'd guess?

[2025-12-27 21:47:18,123]::[InvokeAI]::ERROR --> Error while invoking session 8e1ec091-0679-473b-9a59-8b883e99537b, invocation 2c23b3e7-2fdc-4b7b-aac2-a691d36f
0916 (flux_denoise): 1 validation error for InvocationProgressEvent
percentage           
  Input should be less than or equal to 1 [type=less_than_equal, input_value=1.1666666666666667, input_type=float]
    For further information visit https://errors.pydantic.dev/2.12/v/less_than_equal
[2025-12-27 21:47:18,124]::[InvokeAI]::ERROR --> Traceback (most recent call last):
  File "/home/lstein/Projects/InvokeAI/invokeai/app/services/session_processor/session_processor_default.py", line 130, in run_node
    output = invocation.invoke_internal(context=context, services=self._services)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/lstein/Projects/InvokeAI/invokeai/app/invocations/baseinvocation.py", line 244, in invoke_internal        
    output = self.invoke(context)                                                                                                                                           ^^^^^^^^^^^^^^^^^^^^                                              
  File "/home/lstein/invokeai-main/.venv/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context                               
    return func(*args, **kwargs)                                                                                                                               
           ^^^^^^^^^^^^^^^^^^^^^                                               
  File "/home/lstein/Projects/InvokeAI/invokeai/app/invocations/flux_denoise.py", line 171, in invoke                                                              latents = self._run_diffusion(context)                    
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^                                                                                                                     
  File "/home/lstein/Projects/InvokeAI/invokeai/app/invocations/flux_denoise.py", line 425, in _run_diffusion
    x = denoise(                                                               
        ^^^^^^^^                                                               
  File "/home/lstein/Projects/InvokeAI/invokeai/backend/flux/denoise.py", line 222, in denoise
    step_callback(                                                             
  File "/home/lstein/Projects/InvokeAI/invokeai/app/invocations/flux_denoise.py", line 921, in step_callback
    context.util.flux_step_callback(state)                 
  File "/home/lstein/Projects/InvokeAI/invokeai/app/services/shared/invocation_context.py", line 626, in flux_step_callback
    diffusion_step_callback(                                                   
  File "/home/lstein/Projects/InvokeAI/invokeai/app/util/step_callback.py", line 185, in diffusion_step_callback
    signal_progress("Denoising", percentage, image, (width, height))
  File "/home/lstein/Projects/InvokeAI/invokeai/app/services/shared/invocation_context.py", line 687, in signal_progress
    self._services.events.emit_invocation_progress(
  File "/home/lstein/Projects/InvokeAI/invokeai/app/services/events/events_base.py", line 72, in emit_invocation_progress
    self.dispatch(InvocationProgressEvent.build(queue_item, invocation, message, percentage, image))
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/lstein/Projects/InvokeAI/invokeai/app/services/events/events_common.py", line 149, in build
    return cls(    
           ^^^^                
  File "/home/lstein/invokeai-main/.venv/lib/python3.12/site-packages/pydantic/main.py", line 250, in __init__
    validated_self = self.__pydantic_validator__.validate_python(data, self_instance=self)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
pydantic_core._pydantic_core.ValidationError: 1 validation error for InvocationProgressEvent
percentage
  Input should be less than or equal to 1 [type=less_than_equal, input_value=1.1666666666666667, input_type=float]
    For further information visit https://errors.pydantic.dev/2.12/v/less_than_equal

[lstein]

All the schedulers now work without crashing. Tested on both the linear view and workflow editor mode.

However, the step count does not seem right. For Euler and LCM, when I request six denoising steps I get seven, which is inconsistent with previous behavior. For Heun, I get 11 steps, which is consistent with the behavior in SDXL.

[Pfannkuchensack]

#8705 (comment)

[lstein]

#8705 (comment)

Reply in #8705 (comment).

[lstein]

I'm so sorry, but Euler and LCM are still both reporting N+1 steps when I select N steps. Tested with both Flux Schnell and Flux Dev.

[Pfannkuchensack]

Fixed it here too.

[lstein]

Confirmed working as expected.