mashb1t
commented
8 months ago Please make sure you're using SDXL LoRAs when only using a base model or SD 1.5 LoRAs with a refiner.
https://civitai.com/models/6888/luisap-humanoid-robots-1mb has been trained on SD 1.5
https://civitai.com/models/104976/future-diffusion-robot-youtuber
Read Troubleshoot
[x] I admit that I have read the Troubleshoot before making this issue.
Describe the problem Models (from LORA) seems not to be taken in consideration while generating the result.
Taking this example: https://civitai.com/models/6888/luisap-humanoid-robots-1mb
Trigger words "ROBOT"
I'm using the following model as example:
Style: Fooocus v2, Fooocus Photograph, Fooocus Negative
This is the result with default settings: MODEL: prompt: cyborg girl posing, futuristic robot, robotic girl seed: 8144510124993697307 RealisticStockPhoto_v10 Lora1: SDXL_FILM_PHOTOGRAPHY_STYLE_Betav0.4 = weight 0.25
I get this result, fair enough
Now i'm trying a specific LORA to do this, the one mentioned above prompt: cyborg girl posing, futuristic robot, robotic girl seed: 8144510124993697307 (same) RealisticStockPhoto_v10 Lora1: Cyborgbot = weight 2 (trying lower values also)
no difference... so this clearly means is not taking in consideration the LORA.
Let's try something drastically different, i'm gonna pick a LORA that makes it ANIME, https://civitai.com/models/104976/future-diffusion-robot-youtuber this would be the example expected:
it has the same trigger word "ROBOT" and "CYBORG", so it should work without affecting the prompt.
prompt: cyborg girl posing, futuristic robot, robotic girl (same) seed: 8144510124993697307 (same) RealisticStockPhoto_v10 Lora1: futuredifussion-08 = weight 2 (trying lower values also) with -2 is exactly the same, with 0.5 is the same as well
it's again the same result, is doing nothing with the LORA clearly
I'm gonna try removing the base model, instead of Realistic Stock photo, i'm gonna use the juggernautXL_version6 and again the last LORA model.
Result: Different, but not as per the LORA.
I'm gonna try now removing the STYLE, that I think it might be affecting: So now, no STYLE applied, all empty
Not sure why It changed to this, but still not similar to the LORA:
Testing now removing the SEED; let it be random: Result: even worst:
Let's say the image prompt i'm using, its affecting the result, so let's not use image prompt anymore and let it be just with the text prompt:
I get this; which is not what is expected by the LORA:
So, i'm not sure, what am I doing wrong here? why the loras are not taken in consideration?
Full Console Log
{ "Prompt": "cyborg girl posing, futuristic robot, robotic girl", "Negative Prompt": "unrealistic, saturated, high contrast, big nose, painting, drawing, sketch, cartoon, anime, manga, render, CG, 3d, watermark, signature, label", "Fooocus V2 Expansion": "", "Styles": "[]", "Performance": "Extreme Speed", "Resolution": "(896, 1152)", "Sharpness": 0.0, "Guidance Scale": 1.0, "ADM Guidance": "(1.0, 1.0, 0.0)", "Base Model": "juggernautXL_version6Rundiffusion.safetensors", "Refiner Model": "None", "Refiner Switch": 1.0, "Sampler": "lcm", "Scheduler": "lcm", "Seed": 265211598705829415, "LoRA 1": "futurediffusion-08.safetensors : 2.0", "LoRA 6": "sdxl_lcm_lora.safetensors : 1.0", "Version": "v2.1.862" }