Restore diffusion refinement after 2D upscale

This commit is contained in:
2026-11-12 09:52:10 +00:00
parent 02574168c5
commit b55630c00d
+54 -17
View File
@@ -302,19 +302,18 @@ class DumasH3ContinuumTiledLatentUpscalePerClip:
sampler_select = _native_node("KSamplerSelect")
scheduler = _native_node("BasicScheduler")
is_dumas_2d = isinstance(upscale_param, dict) and upscale_param.get("dumas_h3_2d")
if not is_dumas_2d:
missing = [
name
for name, value in (
("MMH3UltimateUpscale", native),
("RandomNoise", random_noise),
("KSamplerSelect", sampler_select),
("BasicScheduler", scheduler),
)
if value is None
]
if missing:
raise RuntimeError("Missing native ComfyUI nodes: " + ", ".join(missing))
missing = [
name
for name, value in (
("MMH3UltimateUpscale", native),
("RandomNoise", random_noise),
("KSamplerSelect", sampler_select),
("BasicScheduler", scheduler),
)
if value is None
]
if missing:
raise RuntimeError("Missing native ComfyUI nodes: " + ", ".join(missing))
outputs = []
video_outputs = []
@@ -331,7 +330,9 @@ class DumasH3ContinuumTiledLatentUpscalePerClip:
from .dumas_h3_latent_upscale_2d import DumasMMH3LatentUpscale2D
processor = DumasMMH3LatentUpscale2D()
for index, clip_latent in enumerate(latents, start=1):
for index, (clip_latent, clip_model, clip_conditioning) in enumerate(
zip(latents, models, conditionings), start=1
):
samples = clip_latent.get("samples") if isinstance(clip_latent, dict) else None
if samples is None or getattr(samples, "ndim", None) != 5:
raise ValueError(f"clip {index} is not a 5-D H3 LATENT")
@@ -348,10 +349,46 @@ class DumasH3ContinuumTiledLatentUpscalePerClip:
)
audio_latent = audio_values[index - 1]
audio_samples = audio_latent.get("samples")
output = build_av_latent(upscaled["samples"], audio_samples)
nested_latent = build_av_latent(upscaled["samples"], audio_samples)
noise = _unwrap_single_output(
random_noise().get_noise(seed + index - 1), "RandomNoise"
)
sampler = _unwrap_single_output(
sampler_select().get_sampler(sampler_name), "KSamplerSelect"
)
sigmas = sigma_values[index - 1]
if sigmas is None:
sigmas = _unwrap_single_output(
scheduler().get_sigmas(
clip_model, scheduler_name, steps, denoise
),
"BasicScheduler",
)
refine_kwargs = {
"model": clip_model,
"conditioning": clip_conditioning,
"latent": nested_latent,
"noise": noise,
"sampler": sampler,
"sigmas": sigmas,
"cfg": cfg,
# The 2D model has already performed the resize. Passing no
# learned-upscale parameter makes the native node resample
# the target-size tiles only.
"latent_upscale_param": None,
"temporal_split_param": None,
"spatial_split_param": spatial_param,
}
if negatives[index - 1] is not None:
refine_kwargs["negative"] = negatives[index - 1]
refined = _call_native(native, refine_kwargs)
if not isinstance(refined, (tuple, list)) or not refined:
raise RuntimeError(f"native refinement returned no output for clip {index}")
output = refined[0]
outputs.append(output)
video_outputs.append(upscaled)
reports.append(f"clip {index}: {report}")
output_samples, _ = extract_av_streams(output)
video_outputs.append({"samples": output_samples})
reports.append(f"clip {index}: {report}; native tiled refinement applied")
audio_outputs = _clips(audio_latents)
update_geometry = _continuum_helper("update_second_pass_geometry")
updated_plan = update_geometry(