Views
No views yet
/16 variants for 256px generation/16 and /32num_inference_steps=1)FDLossFlowMatchScheduler (scheduler/scheduling_flow_match_fd.py; timesteps t=1→0).
JiT-H/16 FD-SIM at 256×256, 1 NFE, CFG 2.2, interval [0.1, 1.0], seed 42.JiT-B-16-SIM)1from pathlib import Path
2import torch
3from diffusers import DiffusionPipeline
4
5model_dir = Path("./JiT-B-16-SIM")
6pipe = DiffusionPipeline.from_pretrained(
7 str(model_dir),
8 custom_pipeline=str(model_dir / "pipeline.py"),
9 trust_remote_code=True,
10)
11pipe.to("cuda")
12
13generator = torch.Generator(device="cuda").manual_seed(42)
14image = pipe(
15 class_labels="golden retriever",
16 num_inference_steps=1, # 1 NFE (FD-Loss default)
17 guidance_scale=3.0,
18 guidance_interval_min=0.1,
19 guidance_interval_max=1.0,
20 generator=generator,
21).images[0]| Parameter | JiT-B-16-SIM default | Source |
|---|---|---|
num_inference_steps | 1 | --num_sampling_steps 1 |
guidance_scale | 3.0 | JiT_B eval preset |
guidance_interval_min / max | 0.1 / 1.0 | JiT_B eval preset |
legacy_time_convention | True (pipeline default) | --legacy_time_convention |
JiT-L-16-SIM)1model_dir = Path("./JiT-L-16-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6).to("cuda")
7
8image = pipe(
9 class_labels="golden retriever",
10 num_inference_steps=1,
11 guidance_scale=2.4,
12 generator=torch.Generator(device="cuda").manual_seed(42),
13).images[0]JiT-H-16-SIM)1model_dir = Path("./JiT-H-16-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6).to("cuda")
7
8image = pipe(
9 class_labels="golden retriever",
10 num_inference_steps=1,
11 guidance_scale=2.2,
12 guidance_interval_min=0.1,
13 guidance_interval_max=1.0,
14 generator=torch.Generator(device="cuda").manual_seed(42),
15).images[0]| Variant | Path | Architecture | Resolution | CFG (1 NFE) |
|---|---|---|---|---|
| JiT-B-16-SIM | ./JiT-B-16-SIM | JiT-B/16 | 256×256 | 3.0 |
| JiT-L-16-SIM | ./JiT-L-16-SIM | JiT-L/16 | 256×256 | 2.4 |
| JiT-H-16-SIM | ./JiT-H-16-SIM | JiT-H/16 | 256×256 | 2.2 |
| iMF-B-SIM | ./iMF-B-SIM | iMF-B/2 | 256×256 | 8.0 |
| iMF-L-SIM | ./iMF-L-SIM | iMF-L/2 | 256×256 | 8.0 |
| iMF-XL-SIM | ./iMF-XL-SIM | iMF-XL/2 | 256×256 | 8.0 |
| pMF-B-16-SIM | ./pMF-B-16-SIM | pMF-B/16 | 256×256 | 7.5 |
| pMF-B-32-SIM | ./pMF-B-32-SIM | pMF-B/32 | 512×512 | 6.5 |
| pMF-L-16-SIM | ./pMF-L-16-SIM | pMF-L/16 | 256×256 | 7.0 |
| pMF-L-32-SIM | ./pMF-L-32-SIM | pMF-L/32 | 512×512 | 7.5 |
| pMF-H-32-SIM | ./pMF-H-32-SIM | pMF-H/32 | 512×512 | 5.5 |
iMF-B-SIM)IMFPipeline from iMF-diffusers/iMF-B-2 (native iMF time convention, not legacy JiT time). Use torch.float32 (same as base iMF variants):1model_dir = Path("./iMF-B-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6 torch_dtype=torch.float32,
7).to("cuda")
8
9image = pipe(
10 class_labels="golden retriever",
11 num_inference_steps=1,
12 guidance_scale=8.0,
13 guidance_interval_start=0.4,
14 guidance_interval_end=0.65,
15 generator=torch.Generator(device="cuda").manual_seed(42),
16).images[0]iMF-L-SIM)IMFPipeline from iMF-diffusers/iMF-L-2):1model_dir = Path("./iMF-L-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6 torch_dtype=torch.float32,
7).to("cuda")
8
9image = pipe(
10 class_labels="golden retriever",
11 num_inference_steps=1,
12 guidance_scale=8.0,
13 guidance_interval_start=0.4,
14 guidance_interval_end=0.65,
15 generator=torch.Generator(device="cuda").manual_seed(42),
16).images[0]python _convert_imf_l_fd_sim.pyiMF-XL-SIM)IMFPipeline from iMF-diffusers/iMF-XL-2):1model_dir = Path("./iMF-XL-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6 torch_dtype=torch.float32,
7).to("cuda")
8
9image = pipe(
10 class_labels="golden retriever",
11 num_inference_steps=1,
12 guidance_scale=8.0,
13 guidance_interval_start=0.4,
14 guidance_interval_end=0.65,
15 generator=torch.Generator(device="cuda").manual_seed(42),
16).images[0]pMF-B-16-SIM)PMFPipeline from pMF-diffusers/pMF-B-16 (native pMF time convention). Use torch.bfloat16 on CUDA:1model_dir = Path("./pMF-B-16-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6 torch_dtype=torch.bfloat16,
7).to("cuda")
8
9image = pipe(
10 class_labels="golden retriever",
11 num_inference_steps=1,
12 guidance_scale=7.5,
13 guidance_interval_min=0.1,
14 guidance_interval_max=0.8,
15 noise_scale=1.0,
16 generator=torch.Generator(device="cuda").manual_seed(42),
17).images[0]pMF-B-32-SIM)PMFPipeline from pMF-diffusers/pMF-B-32):1model_dir = Path("./pMF-B-32-SIM")
2pipe = DiffusionPipeline.from_pretrained(
3 str(model_dir),
4 custom_pipeline=str(model_dir / "pipeline.py"),
5 trust_remote_code=True,
6 torch_dtype=torch.bfloat16,
7).to("cuda")
8
9image = pipe(
10 class_labels="golden retriever",
11 num_inference_steps=1,
12 guidance_scale=6.5,
13 guidance_interval_min=0.1,
14 guidance_interval_max=0.7,
15 noise_scale=2.0,
16 generator=torch.Generator(device="cuda").manual_seed(42),
17).images[0]