Bili-Sakura/Visual-Generative-Foundation-Model-Collection

A collection of visual generative foundation model including code, paper, checkpoint etc.

stars

commits

Python

primary language

Jun 29, 2026

updated

README

Visual-Generative-Foundation-Model-Collection

[!IMPORTANT] We only hold the core publicly available ones pre-trained on ImageNet.

A selected collection of CORE visual generative foundation model including code, paper, checkpoint etc.

TODO: @src/diffusers Models

Benchmarks

[!NOTE] † DiT-MoE uses additional synthetic training data generated by FLUX and SD3. FID and IS are evaluated on 50k samples, reported with CFG if applicable. ×2 in NFEs indicates that CFG doubles NFEs at inference time.

ImageNet-256

Model	NFE	#Param	GFLOPs	FID	IS	Precision	Recall
Pixel modeling
ADM-G	250×2			4.59		0.82	0.52
PixelFlow		677M		1.98	282.1	0.81	0.60
PixelDiT-XL/16	100×2	700M		1.61
PixelGen-XL/16	50			5.11
JiT-H/16	100×2	953M	182	1.86	303.4
CAFM JiT-H/16	100×2	953M	182	1.80
PixelREPA-H/16		953M	182	1.81	317.2
EPG	1			1.58
PixNerd-XL/16		700M	134	1.93	297
DeCo-XL/16		682M		1.62	301	0.80	0.62
pMF-H/16	1	956M	271	2.22	268.8
Latent modeling
DiT-XL/2	250×2	675M	119	2.27	278.24	0.83	0.57
DiCo-XL/256	250×2	701M		2.05	282.17
DiT-MoE-XL/2-8E2A†		4.1B	323.74	1.72	315.73	0.83	0.64
DiffuSSM-XL-G		673M		2.28	259.13	0.86	0.56
MDT-XL/2		676M	119	1.79	283.01	0.81	0.61
MDTv2-XL/2		676M	119	1.58	314.73	0.79	0.65
FiT-XL/2		824M	153	4.21	254.87	0.84	0.51
SiT-XL/2	250×2	675M	119	2.06	277.50	0.83	0.59
CAFM SiT-XL/2	250×2	675M	119	1.53
SiT-XL/2 + REPA	250×2	675M	119	1.42	305.7	0.80	0.65
SiT-XL/2 + USP		675M	119	7.35	128.50
Self-Flow-XL/2		675M	119	5.70	151.40	0.72	0.67
FiTv2-XL/2		671M	147	2.26	260.95	0.81	0.59
LightningDiT-XL/2		724M	119	1.35	295.3
iMF-XL/2	1	610M	175	1.72	282.0
JLT-B/1	50×2	130M		2.50	232.51
AFM-XL/2	1	675M	119	2.38
AFM-Deep-XL/2	1	675M	119	1.94
LiT-XL/2-G		675M		2.32	265.20	0.82	0.57
RiT
SiT-XL/2 + REG		677M	119	1.36	299.4	0.77	0.66
DDT-XL/2		724M	119	1.26	310.6
DRWKV-H/2		779M	34.95	2.16	275.36	0.83	0.58
NiT-XL		675M	119	2.03	265.26
ProMoE-XL-Flow		1.568B		2.59	265.62
RAE, DiT-DH-XL/2	50×2	1254M	146	1.13	262.6

ImageNet-512

Model	NFE	#Param	GFLOPs	FID	IS	Precision	Recall
Pixel modeling
ADM-G	250×2			7.72		0.87	0.42
PixelDiT-XL/16	100×2	700M		1.81
JiT-H/32	100×2	956M	183	1.94	309.1
EPG				2.35
PixNerd-XL/16		700M	583	2.84	245.6
DeCo-XL/16		682M		2.22	290.0	0.80	0.60
pMF-H/32	1	959M	272	2.48	284.9
Latent modeling
DiT-XL/2	250×2	675M	525	3.04	240.82	0.84	0.54
DiT-MoE-XL/2-8E2A†		4.1B		2.30	298.35	0.85	0.57
DiffuSSM-XL-G		673M		3.41	255.06	0.85	0.49
EDM2-XXL		1523M	552	1.81
SiT-XL/2	250×2	675M	525	2.62	252.21	0.84	0.57
FiTv2-XL/2		671M	525	2.90	263.11	0.83	0.53
LiT-XL/2-G		675M		3.69	207.97	0.85	0.53
DDT-XL/2		724M	525	1.28	305.1
DRWKV-H/2		779M		2.95	265.20	0.84	0.54
NiT-XL		675M	525	1.45	272.77
RAE, DiT-DH-XL/2	50×2	1254M	642	1.13	259.6

Contributors

Bili-Sakura

56 commits

Copilot

5 commits

cursoragent

3 commits

Similar repos

Bili-Sakura/PixNerd-diffusers

[ICLR 2026] PixNerd: Pixel Neural Field Diffusion

shared contributors

Bili-Sakura/EarthBridge-Preview

[2nd place🥈]EarthBridge: A Solution for 4th Multi-modal Aerial View Image Challenge Translation…

shared contributors

nvidia/PixelDiT-ImageNet

Model

wangsssssss/FlowDCN

Model

MCG-NJU/PixNerd-XL-P16-C2I

Model

MCG-NJU/PixNerd

[ICLR 2026] PixNerd: Pixel Neural Field Diffusion

186

MCG-NJU/FlowDCN

[NeurIPS 2024] Exploring DCN-like Architectures for Fast Image Generation with Arbitrary Resolution

aaa-ceku7/GFT

Model

Languages

Python

100.0%

Bili-Sakura/Visual-Generative-Foundation-Model-Collection

A collection of visual generative foundation model including code, paper, checkpoint etc.

stars

commits

Python

primary language

Jun 29, 2026

updated

README

Visual-Generative-Foundation-Model-Collection

[!IMPORTANT] We only hold the core publicly available ones pre-trained on ImageNet.

A selected collection of CORE visual generative foundation model including code, paper, checkpoint etc.

TODO: @src/diffusers Models

Benchmarks

[!NOTE] † DiT-MoE uses additional synthetic training data generated by FLUX and SD3. FID and IS are evaluated on 50k samples, reported with CFG if applicable. ×2 in NFEs indicates that CFG doubles NFEs at inference time.

ImageNet-256

Model	NFE	#Param	GFLOPs	FID	IS	Precision	Recall
Pixel modeling
ADM-G	250×2			4.59		0.82	0.52
PixelFlow		677M		1.98	282.1	0.81	0.60
PixelDiT-XL/16	100×2	700M		1.61
PixelGen-XL/16	50			5.11
JiT-H/16	100×2	953M	182	1.86	303.4
CAFM JiT-H/16	100×2	953M	182	1.80
PixelREPA-H/16		953M	182	1.81	317.2
EPG	1			1.58
PixNerd-XL/16		700M	134	1.93	297
DeCo-XL/16		682M		1.62	301	0.80	0.62
pMF-H/16	1	956M	271	2.22	268.8
Latent modeling
DiT-XL/2	250×2	675M	119	2.27	278.24	0.83	0.57
DiCo-XL/256	250×2	701M		2.05	282.17
DiT-MoE-XL/2-8E2A†		4.1B	323.74	1.72	315.73	0.83	0.64
DiffuSSM-XL-G		673M		2.28	259.13	0.86	0.56
MDT-XL/2		676M	119	1.79	283.01	0.81	0.61
MDTv2-XL/2		676M	119	1.58	314.73	0.79	0.65
FiT-XL/2		824M	153	4.21	254.87	0.84	0.51
SiT-XL/2	250×2	675M	119	2.06	277.50	0.83	0.59
CAFM SiT-XL/2	250×2	675M	119	1.53
SiT-XL/2 + REPA	250×2	675M	119	1.42	305.7	0.80	0.65
SiT-XL/2 + USP		675M	119	7.35	128.50
Self-Flow-XL/2		675M	119	5.70	151.40	0.72	0.67
FiTv2-XL/2		671M	147	2.26	260.95	0.81	0.59
LightningDiT-XL/2		724M	119	1.35	295.3
iMF-XL/2	1	610M	175	1.72	282.0
JLT-B/1	50×2	130M		2.50	232.51
AFM-XL/2	1	675M	119	2.38
AFM-Deep-XL/2	1	675M	119	1.94
LiT-XL/2-G		675M		2.32	265.20	0.82	0.57
RiT
SiT-XL/2 + REG		677M	119	1.36	299.4	0.77	0.66
DDT-XL/2		724M	119	1.26	310.6
DRWKV-H/2		779M	34.95	2.16	275.36	0.83	0.58
NiT-XL		675M	119	2.03	265.26
ProMoE-XL-Flow		1.568B		2.59	265.62
RAE, DiT-DH-XL/2	50×2	1254M	146	1.13	262.6

ImageNet-512

Model	NFE	#Param	GFLOPs	FID	IS	Precision	Recall
Pixel modeling
ADM-G	250×2			7.72		0.87	0.42
PixelDiT-XL/16	100×2	700M		1.81
JiT-H/32	100×2	956M	183	1.94	309.1
EPG				2.35
PixNerd-XL/16		700M	583	2.84	245.6
DeCo-XL/16		682M		2.22	290.0	0.80	0.60
pMF-H/32	1	959M	272	2.48	284.9
Latent modeling
DiT-XL/2	250×2	675M	525	3.04	240.82	0.84	0.54
DiT-MoE-XL/2-8E2A†		4.1B		2.30	298.35	0.85	0.57
DiffuSSM-XL-G		673M		3.41	255.06	0.85	0.49
EDM2-XXL		1523M	552	1.81
SiT-XL/2	250×2	675M	525	2.62	252.21	0.84	0.57
FiTv2-XL/2		671M	525	2.90	263.11	0.83	0.53
LiT-XL/2-G		675M		3.69	207.97	0.85	0.53
DDT-XL/2		724M	525	1.28	305.1
DRWKV-H/2		779M		2.95	265.20	0.84	0.54
NiT-XL		675M	525	1.45	272.77
RAE, DiT-DH-XL/2	50×2	1254M	642	1.13	259.6