<h3 id="v4!">V4!</h3>This version will (probably) be my last version of New Mecha as I am moving to ANIMA. But this promised one is a big rework of the previous version. Using 700 high quality pictures made using <a target="_blank" rel="ugc" href="https://civarchive.com/models/2626223/narnima-turbo-all-in-one-anima-v1-checkpoint">nArnima </a>and then IMG2IMG using New Mecha V3, i did a full finetune of the model to fix several issues:<ul><li>Better hands (that's still Illustrious, but it should work better out of the box)</li><li>Less "unexpected" elements. Less crazyness but better prompt adherence.</li><li>Less saturated (but still very vibrant, i like my reds to pop out).</li><li>Cleaner faces. Some prompts (like cyberpunk) had the tendency to add stuff on skin.</li></ul>This version is slated to replace V3 in the generator (at next turn over, both will still be available), <s>i'll publish a comparative of both to help</s>. Thanks everyone 💖Comparative study available here: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/30664">https://civarchive.com/articles/30664</a><h3 id="v3">V3</h3>This will be the penultimate version of New Mecha. It is still an intermediate step but since it changed a lot from previous version i felt it needed its own number. It is basically Duchesse plus my own "booster" LoRA made from a unexpected finetune (i am actually working on UnNamedIXL V5). The idea is that I extracted a LoRA via SVD of V5 minus V4 and added a bit of it to NewMecha: this results in better backgrounds, some i felt the model was missing. It also add more details (some unexpected/unprompted for), i'll tune this for the final V4 before going away.<h3 id="duchesse">Duchesse</h3>Much like "experimental", this version is a step before V3. And it is a testimony to the incredible work of <a target="_blank" rel="ugc" href="https://civarchive.com/user/Chumple">@Chumple</a> on <a target="_blank" rel="ugc" href="https://civarchive.com/posts/26547351">Duchesse Chaotique</a>!While testing the model, i realised that New Mecha is a but too blurry on background and after a quick test of using my own internal LoRA i made to upgrade New Mecha previously on Duchesse, i though it could integrate a bit of it inside New Mecha to tame it a bit.Here is what the test look like:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1a33069b-f81b-4a6c-89b6-357228663069/original=true/1a33069b-f81b-4a6c-89b6-357228663069.jpeg" />So, i merged this LoRA into duchesse and did a SLERP at low alpha (0.3) of this temporary thingy into V2. This lost a bit of the dark of side of V2, but will constitute a good base to work on V3.NB: For the few people not knowing this regarding merge. When on a model page, look for the Author name and check the icons under it:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d66deb2c-3a42-4ec1-980b-b3d7c7da0016/original=true/d66deb2c-3a42-4ec1-980b-b3d7c7da0016.jpeg" />If this particular icon is here, this is the NOGO for sharing merge (which was not the case for Duchesse Chaotique).<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/60545918-93d6-42a2-80d8-ed78b4d94941/original=true/60545918-93d6-42a2-80d8-ed78b4d94941.jpeg" /><h3 id="v2">V2</h3>Based on Plus after resetting a bit the CLIP and merging back a bit of the first version, it keeps the crazy and excessive style while being more respectful of prompts. It will not "replace" the Initial commit which is more "standard", but i wanted a version more unhinged similar to the now gone Psychonex model.Showcase pictures post contains all the Forge parameters used for generation.<s>PS: we are stuck </s><a target="_blank" rel="ugc" href="https://civarchive.com/changelog?id=129"><s>until pictures can get rating</s></a><s> before the model can be accessed</s><s> 😭</s><s>In any case, i am also waiting for the model to be available on-site at next reset and then, a 2k followers bounty should happen 😉</s><s>Bounty is currently running! </s><a target="_blank" rel="ugc" href="https://civarchive.com/bounties/11070"><s>https://civarchive.com/bounties/11070</s></a>Winning entries are here: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/25315">https://civarchive.com/articles/25315</a><h3 id="plus">Plus</h3>This is not yet the final form of New Mecha, but i wanted to Improve upon Experimental to bring it up to speed on my own "style". The base idea is that i mostly use DPM++ 2M SDE Heun with SGM Uniform as a sampler/scheduler pair, which is not readily available onsite or in some other tools.So, leveraging the dataset i made to make the <a target="_blank" rel="ugc" href="https://civarchive.com/models/2227157/zit-new-mecha-style">ZIT </a>version of New Mecha (which was generated using this pair), i trained a LoRA using Euler A as sampler and used it upon my own checkpoint to help it generate what i wanted with a simpler sampler.This result in a less flat style even with Euler A, which may not fit all intended needs but is closer to my personal taste 😋Usage: 40 steps using Euler A (CFG around 5-6) + a round of img2img at 0.3 denoise should give you similar looking pictures to the showcase one.NB: It got also a bit funky regarding prompt following, which can give nice results but also be a bit frustrating. I'll see to that before a next release (probably with a real fine tune) 😝Also, hands should be fine most of the time, but strange stuff can still happen 😅<h3 id="experimental">Experimental</h3>Long story: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/23172">https://civarchive.com/articles/23172</a><h3 id="dmd2">DMD2</h3>Update DMD2: Merge of the DMD2 fast LoRA, it alter a bit the style but is nice nonetheless (switch to lower CFG, around 1.2, and 6 to 8 steps, including for Hires.fix and Adetailer)<h3 id="new-mecha">New Mecha</h3>I have too many models and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1947875/postbanixl">PostBan </a>was my attempt at creating something new... but i was not satisfied.<ul><li>HoJ is my prefered style, but it is too messy.</li><li>AnBan is nice but too flat.</li><li>Both of them aren't getting me satisfied anymore.</li></ul>That's when i did some tests with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2081281/pleasurechest?modelVersionId=2354947">PleasureChest </a>by @Ereijtic, together with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/1784001?modelVersionId=2023635">UniversalCLIP </a>by @Felldude : by integrating them with my earliest version of HoJ, i was getting something that really like. But the result was broken. Eyes and hands were unusable, i clearly f**ked up something during the merge. That's why, i thought this could be good opportunity to try <a target="_blank" rel="ugc" href="https://github.com/ljleb/sd-mecha">sd-mecha</a>. By leveraging this tool, i was able to ensure my merge is consistent and i can share the recipe.With the additional help of <a target="_blank" rel="ugc" href="https://civarchive.com/models/1772667/indexed">Indexed </a>by @epiTune and <a target="_blank" rel="ugc" href="https://civarchive.com/models/1974650">Artistrious </a>by @Jedas , i feel like i got exactly where i wanted to land (or at least i think so. Before i try to do a V2 and get frustrated again :D)New Mecha will be my new main model going forward. No further versions will be made for AnBan, HoJ, PostBan or other miscellaneous models.Here is the sd-mecha recipe:<pre><code>import sd_mecha as sdm
sdm.set_log_level()

clip_uni = sdm.model("clip\\UniversalCLIPbase.safetensors", config="sdxl-sgm")
clip_v2 = sdm.model("clip\\illus2-clip.safetensors", config="sdxl-sgm")
recipe_clip = sdm.slerp(clip_uni, clip_v2)

recipe_base = (
 sdm.pick_component(sdm.model("illustrious\\pleasurechest_v1.safetensors"), "diffuser") |
 sdm.pick_component(sdm.model("vae\\sdxl_vae_extract.safetensors", config="sdxl-sgm"), "vae") | 
 recipe_clip
)

models = [
 sdm.model("hoj40c.safetensors"),
 sdm.model("shinv3.safetensors"),
 sdm.model("illustrious\\indexed_v1.safetensors"),
 sdm.model("illustrious\\artistrious_v10.safetensors"),
]

recipe_dare = sdm.ties_with_dare(recipe_base,*models,probability=0.6,seed=42,alpha=0.5)

lora = sdm.convert(sdm.model("..\\Lora\\000000\\HoJ4PC.safetensors"), recipe_dare)
recipe_lora = sdm.add_difference(recipe_dare, lora, alpha=0.3)

sdm.merge(recipe_lora, output="NewMecha.safetensors")</code></pre>Some notes if you want to follow along:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/n-Arno/Illustrious-Upgraders/blob/main/UniversalCLIPbase.safetensors">UniversalCLIPbase.safetensors</a> is the merge of CLIP-L and CLIP-G from the link above, with the keys renamed to fit the SDXL expected naming convention (i ended merging it in a very small quantity, together with the Illustrious V2 CLIP)</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/140686/fix-fp16-errors-sdxl-lower-memory-use-sdxl-vae-fp16-fix-by-madebyollin">sdxl_vae_extract.safetensors </a>is the SDXL "FP16 fix" VAE, with the added first_stage_model prefix before the keys for the same reason as above</li><li>HoJ4PC.safetensors is an un-released LoRA i made by substracting HoJ by PleasureChest.</li></ul>This is a messy recipe, my first attempt at using sd-mecha with almost no documentation (i found this <a target="_blank" rel="ugc" href="https://github.com/ljleb/sd-mecha/blob/real-docs/docs/guide/0-introduction.md">doc </a>in an other branch... but it's a little light :D)Regarding usage... well, i think it is un-needed to provide any pointers. Because basically, use the checkpoint as you would any other Illustrious model usually and it should work :DMy usual go-to for local gen is either Euler A or DPM++ 2M SDE Heun and SGM Uniform, 20 or 40 steps depending on the amount of details i want, CFG 4 or 7 depending of the freedom i want to give to the model.Two grid are available in the showcases pictures for demo.You don't really need "quality tags", add the one you like. Sometimes, i just add <a target="_blank" rel="ugc" href="https://civarchive.com/models/1062439?modelVersionId=1192334">IllusP0s</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/1065154?modelVersionId=1470551">SmoothPos </a>or <a target="_blank" rel="ugc" href="https://civarchive.com/models/1220213?modelVersionId=1374643">dfPos</a>. I made a "<a target="_blank" rel="ugc" href="https://civarchive.com/models/2112706?modelVersionId=2390050">MechaP0s</a>" using this model CLIP since it differs from traditional Illustrious a bit.For negative prompt, i usually leverage <a target="_blank" rel="ugc" href="https://civarchive.com/models/1062439?modelVersionId=1192359">IllusN3g</a>, but hey, you can go crazy with your usual neg if you want :DAdetailer is also a good help, but whatever facefix tool you like is fine.Enjoy! (if you do a merge, a small link to here is appreciated.)Disclaimer: This model was posted on Civitai ONLY. If you find this model elsewhere, know that it was not posted by me (but really i don't care anymore, just be aware that commercial use of this model is FORBIDDEN).

(illustrious) New Mecha

Posted on Civitai, upload found elsewhere are done without the author permission.

Files with SHA256: 8974b8f46f596577cbdf2c6c59dbdce158f781e0be872aa3a2ba48bb16e690bd