Hey! This is my first Mix so it might not be the best but i hope you give it a try!
I would love to know what you think of it. All feedback is welcome!
You can try Online Generation Here
TSRMix v7.0 The Evolution Update:
Welcome to the next generation of TSRMix! After many merges, the latest updates began to lean heavily toward a thick, saturated, highly rendered 2.5D style, which was not the objective. This update represents a massive leap forward in versatility, lighting, and overall aesthetic polish. I completely overhauled the model's to improve its performance in style and composition rendering decreasing previous biases. Now TSRMix excels at soft, diffused ambient lighting, beautiful daylight scenes, and gorgeous atmospheric effects like volumetric fog and pastel glows. Linework is thinner, cleaner, and integrates flawlessly into the coloring.
v6.0:
the checkpoint handles fine details more effectively, generates better compositions, and produces facial features with unprecedented quality!
v5:
No LoRAs were used on the showcased images, and most of them are not even upscaled/hires-fixed, so you can achieve high quality without using anything other than prompting.
If you're finding it difficult to perfectly recreate the showcased samples, it's probably because I'm using SwarmUI and the generation process is different from Forge/A1111 (although the output will not be that different).
It is highly recommended to use this model with artist tags, styles tags, or at least stylistic LoRas as the base style is very neutral/simple. It also works fine for both SFW and NSFW.
Recommended Settings:
Sampling Method: DPM++ 2M SDE for v5.0 and DPM++ 2M for v4.0 and below
Scheduler: SGM Uniform
CFG Scale: 3-5
Steps: 25
HirezFix: Remacri, KemonoScale_v2 (thanks to Iragami for letting me know about KemonoScale)
RESOLUTIONS:
832 x 1536
768x1344
832 x 1216
896x1152
1024 x 1024
1152x896
1216 x 832
1344x768
1536 x 832
ADITIONAL OBSERVATIONS:
Some styles i like:
masterpiece, best quality, amazing quality, newest, very aesthetic, absurdres, (scenery:1.3), (by bb \\(baalbuddy\\):0.8), (by rariatto \\(ganguri\\):0.7), (by ilya kuvshinov:1.1)
masterpiece, best quality, amazing quality, newest, very aesthetic, absurdres, (scenery:1.3), (by bb \\(baalbuddy\\)), (by rariatto \\(ganguri\\))
masterpiece, best quality, ultra-detailed, illustration, (graphic novel style), (art by yoji shinkawa:1.2)
masterpiece, best quality,high quality, newest, highres,8K,HDR,absurdres
This mix also have GENESIS Lokr model built on it so you can generate a even wider variety of styles(try adding more weight if you are not seeing much difference).
Here is a list of links of what GENESIS supports:
Artist Expansion:
Nji Style Tags:
Lighting Effect Tags:
Recommended Negative prompt:
lowres, (bad quality,worst quality:1.2), worst detail, sketch, censor, transparent background, patreon logo, bad_anatomy, artist_name, twitter username, patreon username, signature, text
worst quality, bad quality, jpeg, artifacts, (bad hands, bad finger, anatomy), (artist name, copyright name, watermark, text, english text), (bad feet, bad toes), (bad quality,worst quality:1.2), worst detail, sketch, censor, transparent background, patreon logo, bad_anatomy, artist_name, twitter username, patreon username, signature, text, (copyright name:1.2),(watermark:1.2)
Extra:
Shout out to luxiaoaiq233102, EDICT, suliu8, FeR1in, Vigilence, Iragami, Princess_daphiea, Bubbl3gum and Algodoncito. Thanks for your tips!
Description
Improvements in v2.0:
-Higher sharpness, realistic and fine-grained textures;
-Better lighting, shadows and highlights
-Wider range and more harmonious color pallet
-Better with close-ups (it can create amazing images even without hires/upscale/adetailer)
I also managed to tone down the tendency of the model to add artists signature on the images (still can happen but way less. Usually related to the seed)
FAQ
Comments (20)
I am new to these models, illustrous is SDXL based?
Which upscaler do you recommend for HireFix?
I prefer Remacri, you can find it here on CivitAi.
KemonoScale_v2 works best
@Iragami Oh, haven't tryed this one yet.
@ZaeY 1x-ReFocus-V3 - This one also works quite well, but on some checkpoints it adds noise for me, I don't know why...
@Iragami i'll test it out and add to the recomendations on the post, thanks for your feedback :)
This Looks Good... Would Love to try it out!
My account is kinda old but for a long time i only used this site to download models, im not familiarized with how the generation inside civit works. I thought just uploading the model would be enough. Sorry about that I’m still learning how things work here.
Dw. I Generate Locally(Google Colab Cloud). I'll Try it out Later and Post a Review of it
Thank you For Releasing this! This is Great!
V3 Looks Great. If I have Time. I'll Review it!
One of the Best Illustrious/Noob Merges I've Tried out! And V3.0 Just makes it Better. Thank you for Releasing this Model!
@Boredafk Im really happy to know you liked it. Thanks for your feedback and kind words!
Can't Wait for V4.0. But maybe I'll Just Finish Reviewing some Other stuff then I'll Review V3.5. But for now. V3.0 is still my sweetheart. Keep up the Good Work!
@Boredafk Thanks for your feedback! v3.5 is more of a test version so its not as stable as v3.0, but Im working on making it more stable and better aestheticaly for v4.0. I'll release it soon
The V6 Is Reaallllllll!
@Boredafk Yeah haha, I hope you have fun with it!








