Original 1.0.x models are trained on around ~40 images of mecha, robots, machines and power armor using SD 1.5. The images are a mix of drawings, 3D renderings and AI images created in Midjourney.
The 2.0.x variants are trained on 28 Midjourney images of highly photo realistic, sleek robots and armor using SD 2.1 768.
If you download and use this model, PLEASE post a review! I'd value any feedback on the results.
Example prompts:
"+dvMech style, sleek body armor, mecha soldier, portrait, 85mm, f1.8, close up, photo realistic, jungle background, dramatic sunset lighting, shadows, subtle details, diffusion, highlight blooming, low contrast, desaturated"
In the above prompt example, edit the bold italic text above to change the subject, background/environment and lighting.
Here's another example prompt:
"+dvMech style, sleek body armor, mecha soldier, portrait, 85mm, f1.8, close up, photo realistic, desert military base background, soft morning lighting, shadows, subtle details, diffusion, highlight blooming, low contrast, desaturated"
Standard Negative Prompt: pixelated, high contrast, saturated, blurry, oversharpened, sketch, drawing, painting, brush strokes, cutoff head, cropped head, text, type, logo, watermark
All images Made in NMKD GUI, Euler, 25 steps, 768x512, 11.5 CFG, Custom VAE: vae-ft-mse-840000-ema-pruned.ckpt
Description
For this dvMech 2.0.0 custom tuned mode, I did a full new training based on twenty-eight 768x768 images and the SD 2.1 768 base model. Training images were all highly detailed, photo-realistic mech/robots generated by MidJourney.
FAQ
Comments (13)
Is here a config file for the 2.0 version? I can't seem to load it.
Which GUI are you using? I have it in Automatic 1111 and it loads fine.
Ah...another one that only works in the tool least likely to be stable on my machine. ah well. (My tool of choice is InvokeAI)
@imn0mn it's a standard CKPT file, it should load in any SD GUI.
@darkvoyage and yet, it doesn't load in InvokeAI.
I second this; refuses to load with InvokeAI.
>> Loading dvmech_200 from X:/Stable Diffusion/Models/CivitAI/dvmech_200.safetensors
>> Scanning Model: dvmech_200
>> Model scanned ok
>> Loading dvmech_200 from X:/Stable Diffusion/Models/CivitAI/dvmech_200.safetensors
| Forcing garbage collection prior to loading new model
** model dvmech_200 could not be loaded: LatentDiffusion.__init__() missing 1 required positional argument: 'personalization_config'
<stack trace>
TypeError: LatentDiffusion.__init__() missing 1 required positional argument: 'personalization_config'
If I make part robot part human, I can't change the face with any lora or TI
I haven't tried using this model with a LoRA or TI yet, what happens when you try?
RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x128) .How to solve??
Yeah doesn't work for me either. "RuntimeError: output with shape [320, 320] doesn't match the broadcast shape [320, 320, 320, 320]"
It is strange that this fp16 model needs to be cast to fp32 in order to produce output, which is causing many issues
Is that why it's only giving me blurry pixels? I haven't had any issues with this author's other models, but I can't seem to get this one to work properly.
What tutorial did you use to learn?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.