Ruined Fooocus compatibles Flux Checkpoints

[schtroumpf88]

here is a list of working Flux checkpoints from Civitai , i tested this yesterday , and Ruined Fooocus is a pure piece of JOY ;) thanks to the team

Flux aderekFluxFP16Clip_v10.safetensors
Flux 1 CompactCLIPAnd_Flux1DevFp16.safetensors
Flux formidable_v10.safetensors
Flux perfectwoman_v50.safetensors
Flux perfecthumansexy_v50.safetensors
Flux Realistic v1_v10.safetensors
Flux STOIQO Newreality F1D PreAlpha.safetensors
Flux theAraminta_flux1A1.safetensors
Flux FusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO.safetensors
Flux mklan Flux Dev V1 FP8 Clip v1.safetensors

[Yogesh-DevHub]

Wish you long years to live!

[jmc1712]

So far I have only found one as those are the file names and not what they are actually called. Any chance of post actual links as I have found #1 which is not free and #2

[schtroumpf88]

ok

https://civitai.com/models/672766?modelVersionId=753117 Aderek Flux FP16 Clip T5xxl fp8
https://civitai.com/models/637170/flux1-compact-or-clip-and-vae-included CompactCLIPAnd_Flux1DevFp16.safetensors
https://civitai.com/models/679015/formidable formidable 1.0
https://civitai.com/models/320423/perfectwoman perfectwoman 5.0
https://civitai.com/models/313512/perfecthumansexy perfecthumansexy 5.0
https://civitai.com/models/652009?modelVersionId=729439 Flux_Realistic_v1 v 1.0 ( the 2 others models don'y work )
https://civitai.com/models/161068/stoiqo-newreality-or-flux-sd-xl-lightning STOIQO Newreality F1D PreAlpha
https://civitai.com/models/463163?modelVersionId=742904 The Araminta Experiment Flux1-A1 ( version FV1 is for SDXL )
https://civitai.com/models/630820/flux-fusion-ds-smooth-merge-4-steps-aio-and-unet-all-gguf-nf4-fp8fp16 FusionDSSmoothMerge4StepsAIO_v0 Fp8 E4m3fn AIO
https://civitai.com/models/640627/mklan-flux-dev-v1-fp8-clip-vae-included mklan Flux Dev V1 FP8 Clip v1

---

be sure to download the correct version for each one
under each title of a checkpoint name, on Civitai page, there is a list of available versions ( from newest to oldest ) pick the right one
for example perfectwoman has 5 versions ... sometimes the list of availables contains Flux and Sdxl ... so be aware ;)

[jmc1712]

Thanks so much.

[jmc1712]

perfectwoman 5.0 and fluxFusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO are giving an error and not loading. What am I doing wrong?

[jmc1712]

This is the error I get: Loading base model: fluxFusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO.safetensors
ERROR: Error while deserializing header: HeaderTooSmall
Failed to load fluxFusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO.safetensors, loading default model instead

[schtroumpf88]

answers to guys reporting this error is often ' file corrupt . redownload the safetensors file '
my version of this model works without problem

[schtroumpf88]

excellent results with this new one :

https://civitai.com/models/712347/lethaivnhyperrealisticflux LeThAiVN_Hyper_Realistic_Flux v 1.0

[jmc1712]

What is the minimal steps to get a decent output?

[schtroumpf88]

I use a custom in Performance - 12 to 25 steps - cfg 4.5 - euler - simple - with Flux D models

AI Ninja has a couple of great tutos here https://www.youtube.com/watch?v=aCT5RLiPpVs

[jmc1712]

That is where I found out about Ruined. I have tested out 7 models that will run on my system which only has 32 gigs RAM and 4 gigs graphics card. So far I found 3 that crank out nice images using just 3 steps and under 5 minutes. 12 to 25 steps would take too long on my system. I am using cfg 2.5 euler - simple. Below are what I have tested so far that work. *** did nice output with just 3 steps under 5 minutes. all the others will need 10 or more steps to get anything decent which takes longer than I want so I will keep looking for models that can run in 6 or less steps and under 6 minutes.

lux1SFp16_v10 ***
fluxArtfusion4Steps_v12 ***
fluxFusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO
Formidable_v10
mklanFluxDevV1FP8Clip_mklanFluxv1
theAramintaExperiment_flux1A1
flux1DFp16_v10

[jmc1712]

I am getting this error but downlaod had no issues. Failed to load fluxschnellbeasts8ste_v10.safetensors, loading default model instead

Is this a Ruined issue? I have had that happen with a few other Flux models.

[jmc1712]

No problem. Until I can buy a faster PC I will keep testing and find models that can do a good job with 4-6 steps as long as it's under 10 minutes.

[jmc1712]

some Flux models are working in your Fooocus or not ?

Yes. These work but most require 12 or more steps to produce a good image but on my system takes over 12 minutes. I have found 2 that will do nice images in 3-4 steps under 6 minutes. They all work on a 32 gig/4gig system.

lux1SFp16_v10
fluxArtfusion4Steps_v12
fluxFusionDSSmoothMerge4StepsAIO_v0Fp8E4m3fnAIO
Formidable_v10
mklanFluxDevV1FP8Clip_mklanFluxv1
theAramintaExperiment_flux1A1
flux1DFp16_v10

[schtroumpf88]

ok for your Vram
maybe you could test Forge UI ... able to split ram between graphic card and CPU
a very good Flux compatible solution free and unlimited
and Forge is often able to run different Flux checkpoints than Ruined Fooocus
but no idea if it run's on 4 GB vram

https://github.com/lllyasviel/stable-diffusion-webui-forge forge is a creation of the well known genius lllyasviel !! ;)

I'm on 4070Ti 12GB + 32 GB ram

[jmc1712]

Better graphics card is on wishlist.

[Giribot]

Yes, it's run well on a Nvidia 4gb vram but you must check the two options of OOM (left/down in Stable Diffusion WebUI FORGE)

[yownas]

I think I gave you guys a bunch of more work. :)
In the last update RF support Flux gguf-models and models that only contain the unet.

The smallest 2-bit quantized gguf models should work on 6GB vram. The quality of the images aren't the best...but it works.

https://github.com/runew0lf/RuinedFooocus/wiki/Flux-%E2%80%90-GGUF-and-unet-safetensors

[Yogesh-DevHub]

Great work man, Will check it out.

[ckotty-alex]

Thank you so much ! Just tested gguf version of fluxunchained q8, it works very well on 4060ti/16gb, 2m+sgmuniform 1024x1024 20 steps less than a minute

[schtroumpf88]

Tested GGUF or UNET in Ruined Fooocus 1.56

Flux1-dev-f16.gguf : reading of the model is a bit long but the renders are correct and Loras work

Flux1-schnell-F16.gguf : reading of the model is a bit long but the renders are correct and Loras work

projectGaiaFlux1_v10UNET.safetensors ( nice renders )

---

Flux models working in Ruined Fooocus ( 2 )

flux1CompactCLIPAnd_Flux1DevFp16ALT.safetensors ( all 6 models of the Compact serie are working )

mylife_v30.safetensors
fluxDevFp8AnalogArt_v10VersionCheckpoint.safetensors
flux1DevSingleFileT5XXL_v10.safetensors
lethaivnHyper_v10.safetensors ( beautiful results )
ichWilllMEINSTEIL_fluxV10.safetensors ( beautiful results )

[jmc1712]

Thank you. Just installed those and added them to settings.json

[jmc1712]

I did what it said and installed those but I got this still:

Loading base model: acornIsSpinningFLUX_devfp8V11.safetensors
WARNING:root:WARNING: No VAE weights detected, VAE not initalized.
WARNING:root:no CLIP/text encoder weights in checkpoint, the text encoder model will not be loaded.
Failed. Trying to load as Flux unet.

[jmc1712]

Any idea on why I am getting this?

[yownas]

Looks like you put the wrong filename in settings.json. Depending on which clip/t5 models you downloaded you need to match those names in the settings. (We'll probably try to fix this in later versions so it automatically download the files it needs.)

[jmc1712]

Ruined is the one that created the json when loading to model. Will look for an update.

[schtroumpf88]

test compatibility of Ruined Foocus 1.56 with the last bunch of various GGUF, UNET or classic checkpoints ;)

to see the entire list of Civitai Flux models
in Filters - check - Year - Checkpoint - Lora - All - Flux 1.S - Flux 1.D
then https://civitai.com/tag/base%20model and bookmark

acornIsSpinningFLUX_schnellFP8V10.safetensors - reloading of the model at each image generation – bad understanding of prompts
azazealFluxSchnellDev_t5xxlfp16.safetensors – good
azazealFluxSchnellDev_t5xxlfp8.safetensors – very slow – reloading of the model at each image generation
c4pacitor_sV1BetaRC1.safetensors – very slow – reloading of the model at each image generation
bernoulli_q6_K.gguf - very slow – reloading of the model at each image generation
brainflux_v10.safetensors – load as Flux Unet – excellent renders
brainflux_v10GGUFQ80.gguf – excellent renders
c4pacitor_dV1BetaRC1Fp16.safetensors – load as Flux Unet - very slow – reloading of the model at each image generation
centerfoldFlux_v20FP8E5m2.safetensors - load as Flux Unet – poor renders
creapromptFlux1Dev_v10.safetensors – load as Flux Unet – reloading of the model at each image generation
dacFluxedStyle_v10.safetensors – load as Flux Unet - reloading of the model at each image generation – excellent results
daiversityFluxDiverse_v10Fp8Checkpoint.safetensors - load as Flux Unet - reloading of the model at each image generation – excellent results
demonflux_10LowramSchnell.safetensors – good
flux1-dev-F16.gguf – good
flux1-dev-Q5_K_S.gguf – good results
flux1-dev-Q6_K.gguf – good results
flux1-dev-Q8_0.gguf – good results
flux1-schnell-F16.gguf – ok
flux1-schnell.safetensors – ok
flux1SFp16_v10.safetensors - ok
flux1-schnell-Q6_K.gguf – good results
flux1-schnell-Q8_0.gguf – good results
Flux_WoW_3_q4ks.gguf - ok
fluxDevFp8AnalogArt_v10UNET.safetensors – reloading of the model at each image generation – excellent results
fluxheliosbeasts_v10.safetensors – reloading of the model at each image generation – good renders
fluxPureMergeBF16_bf16.safetensors – only noisy pictures very bad renders
fluxRealisticV1_v10NF4FullModel.safetensors – don't work
fluxRealisticV1_v10UNETONLY.safetensors – load as Flux Unet – poor renders
fluxschnellbeasts8ste_v10.safetensors – don't work
fluxUnchainedBySCG_hyfu8StepHybridV10.safetensors – load as Flux unet – reloading of the model at each image generation – good renders
hyperflux1DevBnbNf4_v10.safetensors - don't work
jedpoint_v10.safetensors - don't work
jibMixFlux_v2Fp16_mx5.gguf - don't work
moxieFlux1DS_v128step.safetensors – low quality
pixelwave_flux1Dev01.safetensors – load as Flux unet – reloading of the model at each image generation – good results
pixelwave_flux1Schnell01.safetensors - load as Flux unet – reloading of the model at each image generation – good results
projectGaiaFlux1_v10NF4Fp8T5xxlfp16.safetensors - don't work
projectGaiaFlux1D_v20NF4Uncensored.safetensors – don't work
projectGaiaFlux1_v10UNET.safetensors – load as Flux unet – excellent results
realisticEngineFLUX_devFp8BakedVaeV1.safetensors - don't work
realisticEngineFLUX_devFp8V1.safetensors - don't work
syncadiaFluxV01_fluxV10(1).safetensors – disappointing renders – parts of prompts ignored
topmodel_v40(1).safetensors – excellent results
wowXLPDFlux_fluxWowV4GGUF.gguf – good results

[yownas]

LoRAs should work with GGUF Flux models now. With the caveat that quality could be bad if you use a small, heavily quantized model, like the Q2_K (2-bit) model.

flux1-dev-Q5_K_S.gguf + https://civitai.com/models/790034/horror-dark-film-abandoned-flux

flux1-dev-Q2_K.gguf + https://civitai.com/models/790034/horror-dark-film-abandoned-flux

[CaBaLLaX]

Any chances of adding a Skip button and/or the Stop button be able to stop all prompts at once (if there are multiple prompts separated by ---)?

[schtroumpf88]

another bunch of various GGUF, UNET or classic checkpoints

8StepsCreartHyperFluxDevBnb_hyperDevBnbNf4Unet - don't work
8StepsCreartHyperFluxDevBnb_hyperDevFp8Unet.safetensors – good
8StepsCreartHyperFluxDevBnb_v24HyperFluxDevFp8.safetensors – ok

aderekFlux_fp8.safetensors – load as Flux unet – ok
aderekFlux_v20.safetensors – ok
aderekFlux_withoutClipVae.safetensors – excellent
aderekFlux_fp8V2.safetensors – load as Flux unet – reloading of the model at each image generation – good
aderekFlux_fp8V4.safetensors – load as Flux unet – reloading of the model at each image generation – good
aderekFlux_v5Fp8.safetensors – load as Flux unet – reloading of the model at each image generation – good

acornIsSpinningFLUX_v11Hyper8Step.safetensors – reloading of the model at each image generation – good

agflux_V10Fp8.safetensors – load as Flux unet – reloading of the model at each image generation – good

alternateLowleg_v01PrePreAlphaUNET.safetensors – load as Flux unet – reloading of the model at each image generation – ok
alternateLowleg_v02PreAlphaUNET.safetensors – load as Flux unet – reloading of the model at each image generation – ok

brainflux_v10GGUFQ80.gguf – ok

c4pacitor_dV1 16.safetensors – load as Flux unet – reloading of the model at each image generation – bad results
c4pacitor_dV1 32.safetensors – load as Flux unet – reloading of the model at each image generation – bad results
c4pacitor_dV11.safetensors – load as Flux unet – reloading of the model at each image generation – good

copaxTimelessxl_xplus1.safetensors – excellent
copaxTimelessxl_xplusPoses.safetensors – poor
copaxTimelessxl_xplusUnrealFast.safetensors – good

creapromptFlux1Dev_v10.safetensorsb – load as Flux unet – reloading of the model at each image generation – good

Dev2proFp8SpecialUse_fluxDev2proFp8.safetensors – load as Flux unet – reloading of the model at each image generation – good

DEVDEDISTILLED_fp8.safetensors – load as Flux unet – bad and noisy

epicVision_v1.safetensors - load as Flux unet – reloading of the model at each image generation – good

f1Genuine_v10.safetensors – load as Flux unet – reloading of the model at each image generation – ok

flott_v20810Steps.safetensors – ok

flux1Dev2pro_v10.safetensors – load as Flux unet – reloading of the model at each image generation – ok
flux1DevNF46StepsNSFW_fluxdevFP86Steps.safetensors – load as Flux unet – reloading of the model at each image generation – poor

flux1DevNF46StepsNSFW_fluxdevNF46Steps.safetensors – don't work
flux1DevNF46StepsNSFW2_fluxdevNF46Steps.safetensors – don't work
flux1DevNF46StepsNSFW_hybridFP84StepsNSFW.safetensors – load as Flux unet – poor
flux1DevNF46StepsNSFW2_fluxdevFP86Steps.safetensors – load as Flux unet – reloading of the model at each image generation – poor

flux1DPDXLConsistency_v10VAE.safetensors – poor

flux1RatioMerges_101.safetensors – don't work

flux1-schnell-Q5_0.gguf – good

flux_dev 16.safetensors – good
flux_dev 32.safetensors – good

fluxFusionV24StepsAIOUNETALL_V2Fp16.safetensors – poor

fluxHyperDev8StepsDEV_hyperDevModel10.safetensors - load as Flux unet – good

fluxPureMergeBF16_bf16.safetensors – ok

fluxRealisticV1_ggufFluxRealisticV1.gguf – poor

flux unchainedNF4_fluxunchainedV11NF4.safetensors – don't work

genuine_v20CkptFp8.safetensors – excellent

iaPLAY_v10.safetensors – ok

jibMixFlux_fp8V2Electric.safetensors – load as Flux unet – reloading of the model at each image generation – ok

mangledMergeFlux_v0Bfloat16Dev.safetensors – load as Flux unet – reloading of the model at each image generation – excellent
mangledMergeFlux_v0Bfloat16Schnell.safetensors – bad
mangledMergeFlux_v0Q80DevGGUF.gguf – excellent

matrix_v10.safetensors – good

moxieFlux1DS_v123.safetensors – ok – bad hands
moxieFlux1DS_v13.safetensors – excellent
moxieFlux1DS_v138Step.safetensors – poor

moxieFusionFlux_v1338step.safetensors – poor
moxieFusionFlux_v1328step.safetensors – ok
moxieFusionFluxF16_v1328step.safetensors – ok
moxieFusionFluxNF4_v1348step.safetensors – don't work

nepotismFUXGGUFMeA_v1AIOFP16.safetensors – excellent
nepotismFUXGGUFMeA_v1AIOFP8.safetensors – excellent
nepotismFUXGGUFMeA_v2AIO.safetensors – excellent
nepotismFUXGGUFMeA_v3AIO.safetensors – excellent
nepotismFUXGGUFMeA_v4AIO.safetensors – excellent
nepotismFUXGGUFMeA_v4Q8.gguf – ok
nepotismFUXGGUFMeA_v5Q4KS.gguf – good
nepotismFUXGGUFMeA_v5Q8.gguf – good
nepotismFUXGGUFMeA_v6Dit.safetensors – load as Flux unet – reloading of the model at each image generation – poor

newton_full.safetensors- don't work
newton_unetOnly.safetensors – don't work

nsfwMASTERFLUXFp8LoraMerged_v11.safetensors – load as Flux unet – reloading of the model at each image generation – excellent

projectGaiaFlux1D_v20NF4UNETUncensored.safetensors – don't work

projectHadesFAST_v1NF4AllInOne.safetensors – don't work

projectHadesFLUXS_v11NF4AllInOne.safetensors – don't work
projectHadesFLUXS_v11UnetOnly.safetensors – load as Flux unet – reloading of the model at each image generation – ok

precizeFluxDevNF4FP8FP16_v10Fp8.safetensors – don't work
precizeFluxDevNF4FP8FP16_v10Fp16 – don't work

realFascination_v10.safetensors – load as Flux unet – reloading of the model at each image generation – excellent

realflux10b_10bTransformer.safetensors – low quality
realflux10b_10bCompactSchnell.safetensors – low quality
realflux10b_10bTransformerDev.safetensors – good

realHornyProUnetNF4_realHornyProUnet.safetensors – load as Flux unet – reloading of the model at each image generation – excellent

realisticDeepdream_fluxDCLIPVAEBaked.safetensors – excellent

schnell_v11.safetensors – very good

STOIQONewrealityFLUXSD_F1DAlpha.safetensors – load as Flux unet – ok
STOIQO_Afrodite_v1.safetensors – don't work
STOIQOAfroditeFLUXXL_F1DAlpha.safetensors – good

titanforgeUncensor_titanforgeV1FP8.safetensors – excellent

visionRealistic_v1FluxDevFp8.safetensors – load as Flux unet – reloading of the model at each image generation – ok
visionRealistic_v2FluxDevFp8.safetensors – load as Flux unet – reloading of the model at each image generation – good

vnsDreamWorldMixFlux_v10.safetensors – poor renders