Refresh optimized workflow baselines and MiniMax API graph

This commit is contained in:
Morpheus
2026-09-28 07:13:25 +00:00
parent 4e6af2d813
commit bf074116b9
5 changed files with 58 additions and 6 deletions
+5
View File
@@ -4,4 +4,9 @@ Use [`flux/`](flux/) or [`qwen/`](qwen/) for one-reference-image transformations
The shared reference asset and static variables are recorded in [`variables.json`](variables.json) and `character_reference.png`. The test plans live in each model README. The shared reference asset and static variables are recorded in [`variables.json`](variables.json) and `character_reference.png`. The test plans live in each model README.
## Current best
Qwen I2I: [`qwen/`](qwen/) — Q-I2I-34, Euler CFG++, beta scheduler, 50 steps,
CFG 3.5, approximately 68.3 seconds.
Store each I2I test under the selected model's `results/<test-id>.png` and `results/<test-id>.json` using the table ID as the label. Store each I2I test under the selected model's `results/<test-id>.png` and `results/<test-id>.json` using the table ID as the label.
+6 -2
View File
@@ -23,9 +23,10 @@ The workflow uses the installed Qwen Image 2.1 stack and the flattened `images.i
- `workflow.json` — API-format workflow. - `workflow.json` — API-format workflow.
- `ui-workflow.json` — native ComfyUI web-editor workflow. Load this file in ComfyUI for an editable graph. - `ui-workflow.json` — native ComfyUI web-editor workflow. Load this file in ComfyUI for an editable graph.
## Planned optimization matrix ## Optimization matrix
All rows are intentionally **Untested**. Keep the reference image, prompt, seed, and dimensions from [`../variables.json`](../variables.json) fixed. All 34 planned rows have been tested. Keep the reference image, prompt, seed,
and dimensions from [`../variables.json`](../variables.json) fixed.
| ID | CFG | Steps | Sampler | Scheduler | Runtime (s) | Tested? | Current Best | | ID | CFG | Steps | Sampler | Scheduler | Runtime (s) | Tested? | Current Best |
|---|---:|---:|---|---|---:|---|---| |---|---:|---:|---|---|---:|---|---|
@@ -73,3 +74,6 @@ All rows are intentionally **Untested**. Keep the reference image, prompt, seed,
- Q-I2I-01 remains a strong reference baseline, but its reported 0.23s matrix runtime was cached and is not comparable. - Q-I2I-01 remains a strong reference baseline, but its reported 0.23s matrix runtime was cached and is not comparable.
- Q-I2I-20–23 were valid alternatives; the non-CFG++/ancestral variants did not improve fidelity. Q-I2I-24–26 were rejected as corrupted outputs. - Q-I2I-20–23 were valid alternatives; the non-CFG++/ancestral variants did not improve fidelity. Q-I2I-24–26 were rejected as corrupted outputs.
- Q-I2I-27–31 were valid scheduler comparisons, but none surpassed beta. Q-I2I-02–19 showed no better combination than the CFG 3.5/high-step result. - Q-I2I-27–31 were valid scheduler comparisons, but none surpassed beta. Q-I2I-02–19 showed no better combination than the CFG 3.5/high-step result.
All rows are complete. Q-I2I-24–26 are retained as tested but rejected for
corrupted output; Q-I2I-34 is the sole Current Best.
+3
View File
@@ -5,3 +5,6 @@ duration, and CFG variables for like-for-like comparisons. Efficiency nodes
are tested first on 3-second 480p clips before longer generations. are tested first on 3-second 480p clips before longer generations.
- [`minimax/`](minimax/) — MiniMax H3 R2V base and efficiency matrix - [`minimax/`](minimax/) — MiniMax H3 R2V base and efficiency matrix
Current speed baseline: MiniMax MM-R2V-51, Turbo LoRA, 8 steps, Euler,
simple scheduler, 864×480, 3 seconds.
+16 -4
View File
@@ -1,8 +1,9 @@
# MiniMax H3 — optimized reference-to-video # MiniMax H3 — optimized reference-to-video
This is the MiniMax H3 R2V evaluation workflow. It is derived from the This is the MiniMax H3 R2V evaluation workflow. The current production
imported MiniMax H3 int8 graph and kept separate from the source workflows baseline is the validated Turbo LoRA graph in `api-workflow.json`; the older
under `models/minimax/`. web-editor graph remains in `workflow.json` for reference and is not the
current baseline.
## Fixed comparison inputs ## Fixed comparison inputs
@@ -23,11 +24,18 @@ substitute another configuration.
## Files ## Files
- `workflow.json` — ComfyUI web workflow derived from the source graph - `workflow.json` — ComfyUI web workflow derived from the source graph
- `api-workflow.json` — validated API graph for the current Turbo baseline
- `reference.png` — fixed character reference - `reference.png` — fixed character reference
- `prompt.txt` — fixed haunted-room investigation prompt - `prompt.txt` — fixed haunted-room investigation prompt
- `variables.json` — fixed comparison variables - `variables.json` — fixed comparison variables
- `results/` — labelled outputs and matching JSON metadata - `results/` — labelled outputs and matching JSON metadata
## Current best
**MM-R2V-51** is the current speed baseline: Turbo LoRA enabled, 8 steps,
Euler sampler, simple scheduler, CFG 1, 864×480, 3 seconds, and 381.240s
runtime. MM-R2V-52 is the 12-step comparison at 471.862s.
## Test matrix ## Test matrix
All rows are planned; only the scheduled MM-R2V-01 and MM-R2V-02 control/ All rows are planned; only the scheduled MM-R2V-01 and MM-R2V-02 control/
@@ -86,10 +94,14 @@ results and should not be queued automatically with Phase 1.
| MM-R2V-48 | H quality sweep | Base/Sol-Attn path | 20 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — | | MM-R2V-48 | H quality sweep | Base/Sol-Attn path | 20 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — |
| MM-R2V-49 | H quality sweep | Base/Sol-Attn path | 24 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — | | MM-R2V-49 | H quality sweep | Base/Sol-Attn path | 24 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — |
| MM-R2V-50 | H quality sweep | Base/Sol-Attn path | 28 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — | | MM-R2V-50 | H quality sweep | Base/Sol-Attn path | 28 | er_sde | simple | 3 | 1.0 | Blocked | — | No | — |
| MM-R2V-51 | H Turbo sweep | Turbo LoRA only | 8 | euler | simple | 3 | 1.0 | Yes | Yes | No | 381.240 | | MM-R2V-51 | H Turbo sweep | Turbo LoRA only | 8 | euler | simple | 3 | 1.0 | Yes | Yes | Yes | 381.240 |
| MM-R2V-52 | H Turbo sweep | Turbo LoRA only | 12 | euler | simple | 3 | 1.0 | Yes | Yes | No | 471.862 | | MM-R2V-52 | H Turbo sweep | Turbo LoRA only | 12 | euler | simple | 3 | 1.0 | Yes | Yes | No | 471.862 |
| MM-R2V-53 | H Turbo sweep | Spectrum + Turbo LoRA | 8 | euler | simple | 3 | 1.0 | Blocked | — | No | — | | MM-R2V-53 | H Turbo sweep | Spectrum + Turbo LoRA | 8 | euler | simple | 3 | 1.0 | Blocked | — | No | — |
| MM-R2V-54 | H Turbo sweep | Spectrum + Turbo LoRA | 12 | euler | simple | 3 | 1.0 | Blocked | — | No | — | | MM-R2V-54 | H Turbo sweep | Spectrum + Turbo LoRA | 12 | euler | simple | 3 | 1.0 | Blocked | — | No | — |
| MM-R2V-55 | I Turbo sampler/scheduler | Turbo LoRA only | 8 | euler | beta | 3 | 1.0 | Queued | — | No | — |
| MM-R2V-56 | I Turbo sampler/scheduler | Turbo LoRA only | 8 | euler_cfg_pp | simple | 3 | 1.0 | Queued | — | No | — |
| MM-R2V-57 | I Turbo sampler/scheduler | Turbo LoRA only | 8 | res_multistep | beta | 3 | 1.0 | Queued | — | No | — |
| MM-R2V-58 | I Turbo sampler/scheduler | Turbo LoRA only | 8 | res_multistep | simple | 3 | 1.0 | Queued | — | No | — |
## Result requirements ## Result requirements
+28
View File
@@ -0,0 +1,28 @@
{
"92": {"inputs":{"video":["130",0],"filename_prefix":"video/MiniMax_H3","format":"auto","codec":"auto"},"class_type":"SaveVideo"},
"115": {"inputs":{"aspect_ratio":"16:9 (Widescreen)","megapixels":0.4,"multiple":32},"class_type":"ResolutionSelector"},
"119": {"inputs":{"vae_name":"minimax_h3_video_vae_int8_convrot.safetensors"},"class_type":"VAELoader"},
"120": {"inputs":{"vae_name":"minimax_h3_audio_vae_fp32.safetensors"},"class_type":"VAELoader"},
"121": {"inputs":{"samples":["125",0],"vae":["120",0]},"class_type":"VAEDecodeAudio"},
"122": {"inputs":{"samples":["125",0],"vae":["119",0]},"class_type":"VAEDecode"},
"123": {"inputs":{"sampler_name":"euler"},"class_type":"KSamplerSelect"},
"124": {"inputs":{"model":["127",0],"steps":8,"scheduler":"simple","denoise":1},"class_type":"BasicScheduler"},
"125": {"inputs":{"noise":["129",0],"guider":["126",0],"sampler":["123",0],"sigmas":["124",0],"latent_image":["136",1]},"class_type":"SamplerCustomAdvanced"},
"126": {"inputs":{"model":["141",0],"conditioning":["136",0]},"class_type":"BasicGuider"},
"127": {"inputs":{"unet_name":"minimax_h3_ref2va_pruned_int8_convrot.safetensors","weight_dtype":"default"},"class_type":"UNETLoader"},
"128": {"inputs":{"clip_name":"qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors","type":"minimax","device":"default"},"class_type":"CLIPLoader"},
"129": {"inputs":{"noise_seed":20260927},"class_type":"RandomNoise"},
"130": {"inputs":{"images":["122",0],"audio":["121",0],"fps":24,"bit_depth":8,"color_space":"sRGB","codec":"none","filename_prefix":"MiniMax/MM-R2V-51"},"class_type":"CreateVideo"},
"131": {"inputs":{"values.a":["132",0],"expression":"max(5, round(a * 24)) + (5 - (max(5, round(a * 24)) % 17)) % 17"},"class_type":"ComfyMathExpression"},
"132": {"inputs":{"value":3},"class_type":"PrimitiveFloat","_meta":{"title":"Float (Duration)"}},
"136": {"inputs":{"clip":["128",0],"vae":["119",0],"audio_vae":["120",0],"ref_images.ref_image_0":["137",0],"ref_images.ref_image_1":["139",0],"prompt":["138",0],"width":864,"height":480,"length":75,"ref_image_size":"match"},"class_type":"MiniMaxH3ReferenceToVideo"},
"137": {"inputs":{"image":"reference.png"},"class_type":"LoadImage"},
"138": {"inputs":{"value":"Use <Picture 1> as the visual reference. An adult man investigates a haunted room, cautiously moving through the dark abandoned interior with a flashlight, subtle supernatural movement in the shadows, cinematic suspense, realistic motion and coherent identity."},"class_type":"PrimitiveStringMultiline","_meta":{"title":"Input Text (Prompt)"}},
"139": {"inputs":{"image":"reference.png"},"class_type":"LoadImage"},
"141": {"inputs":{"on_false":["127",0],"on_true":["145",0],"switch":["146",0]},"class_type":"ComfySwitchNode","_meta":{"title":"If/Else Switch (model)"}},
"142": {"inputs":{"on_false":["143",0],"on_true":["144",0],"switch":["146",0]},"class_type":"ComfySwitchNode","_meta":{"title":"If/Else Switch (Steps)"}},
"143": {"inputs":{"value":20},"class_type":"PrimitiveInt","_meta":{"title":"Int (Full)"}},
"144": {"inputs":{"value":4},"class_type":"PrimitiveInt","_meta":{"title":"Int (Lightning LoRA)"}},
"145": {"inputs":{"model":["127",0],"lora_name":"minimax\\KJminimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors","strength_model":1},"class_type":"LoraLoaderModelOnly"},
"146": {"inputs":{"value":true},"class_type":"PrimitiveBoolean","_meta":{"title":"Boolean (Enable Lightning LoRA)"}}
}