Fenris ( Fenrisúlfr) , Fenrir or also old norse "Vánagandr" , the First son of Loki it is foretold that he will kill Odin in Ragnarok.....
This model was actually made for animals and creatures but with the further progress of the model it's also good as an general model for all sorts of generations.
Changelog / Versions details / Credits
V1.4
First Released Model very outdated!
V1.8
Merged in
XL6 - HEPHAISTOS and some loras
did get more training from images by the fenris (1.5) model (this time from 3.1 and 4.03)
overall a better look than 1.4 but still in an early stage (WIP)
V3.1
1 week more training (+100 Pictures)
merged in v1.1 XL6 - HEPHAISTOS
Overall the best Version so far.
V4.4
pulled back because of a bug
V4.8
More training
now have we left the early stage but its still (WIP)
merged in V2.1 of XL6 HEPHAISTOS and IcedCofeexl
Please Check out their amazing work!
V5.01
added more training
better overall quality
V5.5
+more training ( changed also a bit my training Method)
merged in small amounts of following models NewdawnXL, FurtasticXl
Please Check them also out
Note:
The best results I got from CFG 1 to 6 at 40 to 80 steps
7-11 CFG 70 to 200 steps (for example 10CFG /100-150 steps)
or at 12-16 CFG (200-300 at 12) / 400 - 600 steps at 16 CFG
(
DPM++ SDE)V7.0 / V7.03
refined V5.5 also merged with models (6.xx) that never get released
merged in XL13 Asmodeus
I decided to split the model thats the reason for two V7.0 Versions
V7.0
for creative artwork, anime , comic, drawings, and digital art
7.03
if you need more realistic outcomes.
Note: for 7.0/7.03 i recommend using the "FenrisXL VAE
for the next Versions i will bake this VAE in from the start!
V8.01
added more training
merged in small amounts of XL6 - HEPHAISTOS V 3.3
trained in small amounts of ashrpg loras
more realistic, consistent and creative than V7.03
it can handle a huge range of CFG from 1-30+
Please Check out their amazing work
V8.09
Merge between V8.01 and V7.00, V7.02 ,V5.01. , NewdawnV2.2
I also added a bit train difference from Yamers Perfect Design
Model
Made for creative stuff (Digital Art, anime, and drawings)
i recommend using V9.22 for photorealistic Images.
V9.22
Merged in some small parts of the Thalos Project,
Some parts of Fenrisxl 8.09
and lots of new Training Data!
better overall quality
a bit more creative than V8.01 but for art i recommend V8.09
Version got deleted because it had a bug that let the model overbake sometimes and ruined the image sorry about that
V10.03
Fixed V9.22
Recommended for Photorealism
V11.12 (Creative Photo)
Brings the design back from V8.01 (i also merged in back i huge amount of it back in)
added back the posibility for creative stuff did archieved that because i merged in a small amount of newdawnXL
merged also in Colossus Project XL v1.0
i also switched the merging method a bit thanks to Triple_Headed_Monkey
there will be no creative artwork Version anymore because i still recommend using NewdawnXL or V8.09 for that
V12.8 (Photorealistic)
This Model is the upgrade of V10.03
More photorealistic
more training overall
I recomment for Creative Photos V11.12
V14.5 Experimental Version
this is basically Newdawnxl with an fenrisxl picture output
more linguistic and intuitive
more creative
can go from low resolutions from 256x256 (512x512 recommended) to 1200x1200 but can go over this limit too but no guarantee for consistency anymore
it can do artworks again but i still recommend newdawnxl for that
There is also an tensor.art only version with an slightly alternative output and some small technical improvements over 14.50 I recommend this version to use online when you have an potato computer or you like generating images on the go
https://tensor.art/models/645967964966858122
V15.8
It been a while since the last update because it was pretty hard to improve FenrisXL
Merged in Gleipnir
and add lots of Extra Training:
Visual improvements
More versatile and an increased creativity include artworks and drawings but i recommend also looking into Newdawnxl
V16.4
I used some parts of my Bifröst Project model to update this Model this time
more Photorealistic
more consistent
more flexible
Even the results are good I'm always working on the model to make it even better in some way expect lots of updates, experimental stuff ,I'm very creative and have lots of ideas and like trying to break technical limits. That's the reason I'm never done with this model.
FenrisXL [SDXL Lightning]
V1.5 First released Version
This is the First SDXL Lightning Version of Fenris XL and still experimental:
if you need infos what SDXL Lightning is: you can read here about it:
very fast model
very detailed
better in art styles
no lora needed for sdxl [Lightning]
for settings i recommend the following settings:
dpm++ sde and sgm uniform scheduler between 4-15 steps and and around 2-max 4 CFG
(i use for myself most of the time 7 steps )
( see example images)
Note; not every sampler are working: Here a list of working ones:
euler, euler_a, dpm_2_ancestral, dpm_adaptive, dpmpp_sde, dpmpp_sde_gpu,
ddim,uni_pc_bh2, for more infos i made an xyz grid i will upload it soon.
for creatures i can highly recommend to use the Furry Enhancer
for a workflow you can use the Fenrisxl Workflow made by Afroman4peace
this model version use parts of following models , Hephaistos Next Gen thanks for your help Afroman4peace, parts of Ratatoskr and of course parts of the original SDXL Lightning Model
V18
more precision
more styles
without changing the classic fenrisxl style
Added elements of my model RatatoskrV7 and Ragnarök inside the model.
for Release FP32 / Bf16 is available, the FP16 version will follow later.
for the workflow you can check out
i recommend for the workflow to use for example the ragnarök workflow
but i will do a full sampler / cfg grid soon on this model soon too.
please check out also out the newer models like Ragnarök for example.
Fenris Flux V1
First Version of the Flux Variant of Fenris this model is based on Ratatoskr Flux (V1)
This model has an Style Lora inside trained on 1200 output Pictures generated by Fenrisxl
(15,8-V18) i recommend the following workflow
I can recommend for settings: 10 steps 3.5 cfg and euler/simple or ddim/ beta
Fenris Flux V2.3
Finally after Month in training i am Done With the new Version of Fenris Flux.
This Version is trained on a mix of Fenris V1 / Helheim Flux Base Model so its an hybrid model.
it also contains an custom trained Clip_L made for this model so i strongly recommend using it in comfy instead of a separate dual clip ( see my workflows embedded in all Gallery pictures)
more realistic
but can still do art
trained on extended database compared to V1
recommended settings to start with: 16 Steps / Sampler Deis / Scheduler beta
Have fun with the new Model and please give a thumbs up to support my work
Note i will also release the fp16 and the gguf variant stay tuned
please give some feedback if you like ;-) to improve it even further
and if you like buy my buddy a beer that we will then drink together
https://ko-fi.com/afroman4peace
!!!Use it at your own risk, you a responsible for your own creations!!!
important: for the Flux variant: I got asked if there are parts of dev1 in it the answer is yes a tiny amount if you consider it for commercial use do it on your own risk!
PS: will report every sick photo that is posted here! (CP, brutal Gore, etc)
This disclaimer is not here without reasons. please have in mind this is like a virtual showroom and posting extreme stuff is like 💩 on my desk in real live 😉
made by Freek22
Description
Experimental Version!
this is basically Newdawnxl with an fenrisxl picture output
more linguistic and intuitive
more creative
can go from low resolutions from 256x256 (512x512 recommended) to 1200x1200 and even higher but with no guarantee of consistency 😉
No fixed resolutions ( theoretically)
it can do artworks again but i still recommend newdawnxl for that
FAQ
Comments (12)
Looks like the latest version is a bf16 model (I'm on MPS and I can't run those)- any chance of getting an fp16 version? Thanks!
the 14.5? no its an "fp16" but experimental one
I will upload it also on the weekend to tensor than everyone can use it online for free
Hmm, I got "TypeError: Trying to convert BFloat16 to the MPS backend but it does not have support for that dtype." when trying to run it.
@billthekobold very strange but possible that it's not recognised correctly this is an experimental model .....I have an💡 idea I put this model tomorrow in an converter and (when) the model still works I upload it on tensor to my second account. I will post the link here when it's done
Sounds great, thanks!
@billthekobold i have convert this version the output did change a bit ( sometimes even better than V14.5) https://tensor.art/models/645967964966858122
@freek22 Cool, I'll check it out
Thanks for making an SDXL model that allows smaller image renders!
This model runs on my old 750Ti which has only 2gb vram :-)
The results are looking really beautiful!
Are there more SDXL models capable of rendering 512x512 that i should check out?
i can mention https://civitai.com/models/119279?modelVersionId=159123 here . Fenrisxl did build its fundaments on this model
and it works only with 2gb Vram? its huge that it runs on that low specs iam impressed ;-)
@freek22 yeah it does, it took a while of trying all of auto1111's optimizations to find a combination to work even with SD 1.5, but for what it is this ancient GPU is a legend. Can't wait to grab a 12GB 3060 and finally get "decent" render times.
So far your model and the Hephaistos which you recommended are the only ones that work with the GPU, and it's only because the minimum "base size" for the others is 1024x1024 which doesn't fit into my vram.... so thanks again for a chance to get started with SDXL 🧡

















