huehuehue9001 Posted January 20 Posted January 20 (edited) Did you make game UI lora? I would just use screenshot and inpainting the characters. Owner-kun, you gave me the best birthday present. Edited January 20 by huehuehue9001 4
5250 Posted January 20 Posted January 20 (edited) 5 hours ago, huehuehue9001 said: Did you make game UI lora? I would just use screenshot and inpainting the characters. Owner-kun, you gave me the best birthday present. I made Mod to delete Character and delete background and delete foreground in Game. and I got background screenshot and foreground screenshot. Then I used [LayerDiffuse] in Sitable Diffusion. this way is not clear. So your way is best! I will try to use inpainting. Thank you. Edited January 20 by 5250 2
huehuehue9001 Posted January 20 Posted January 20 Your way is best. Inpainting is easier but sometimes it doesn't blend the transition well if the style is too different. 1
5250 Posted January 24 Posted January 24 (edited) Still I can't get Azusa...... Generated same prompt used wildcards. Edited January 24 by 5250 7
5250 Posted January 29 Posted January 29 (edited) His DOAXVV cloth Loras is too wonderful. https://civitai.com/user/gen558/models?tag=clothing Edited January 29 by 5250 3
huehuehue9001 Posted February 4 Posted February 4 (edited) anime3dxl + doaxvv lora. 1MP t2i then upscaled to 1440p Loaded image into Qwen image edit 2509. Upscaled to 4k. denoise 0.4. Without any character lora, qwen really have that same face look. The background is good, the depth of field is smooth. Upscaling to 4k with anime3dxl is not good. The background is all grainy. Edited February 4 by huehuehue9001 5
huehuehue9001 Posted February 7 Posted February 7 (edited) bacon? beef jerky? It's not her birthday. Just trying out stuff. so it's anime3dxl + lora > qwen upscale> inpainting back with anime3dxl + lora maybe i'll try z image upscale. Here is some testing. anime3dxl 4k upscale. (2 minutes to gen) Spoiler flux2 4b (9 minutes) Spoiler z image. (1 minute) Spoiler base upscale (base image is 1344x768, then upscaled to 2528x1440) this step is because 1MP is too low res to get a good face. Spoiler qwen took 4 minutes i think. Spoiler anime3dxl is grainy. flux2 4b looks ok but it doesn't look 4K. zimage is blurry and soft. qwen looks 4K but the face is changed, nipples and cunny is baloney. so you have to fix it. Edited February 7 by huehuehue9001 6
5250 Posted February 8 Posted February 8 (edited) 21 hours ago, huehuehue9001 said: bacon? beef jerky? It's not her birthday. Just trying out stuff. so it's anime3dxl + lora > qwen upscale> inpainting back with anime3dxl + lora maybe i'll try z image upscale. Here is some testing. anime3dxl 4k upscale. (2 minutes to gen) Reveal hidden contents flux2 4b (9 minutes) Reveal hidden contents z image. (1 minute) Reveal hidden contents base upscale (base image is 1344x768, then upscaled to 2528x1440) this step is because 1MP is too low res to get a good face. Reveal hidden contents qwen took 4 minutes i think. Reveal hidden contents anime3dxl is grainy. flux2 4b looks ok but it doesn't look 4K. zimage is blurry and soft. qwen looks 4K but the face is changed, nipples and cunny is baloney. so you have to fix it. I tried to use Nova anime3dxl and others check point on Stable Diffusion Forge. I don't use upscaler, because it slow down the generation speed. I use only face detailer. So to make each pic need about 17 seconds. Since the resolution is reduced when the image is animated, I felt there was no need to upscale it. So I made 1200-1600 dot pic on Forge.(Non upscale) then This pic is down scale (480-640) on ConfyUI, Then Upscale to 960-1280. I'm also trying out a lot of different things. novaAnimeXL_ilV160 nova3DCGXL_ilV80 novaOrangeXL_exV08 novaAsianXL_illustriousV70 Edited February 8 by 5250 3
huehuehue9001 Posted February 8 Posted February 8 (edited) the nova anime looks closest to game. you can also try uncanny valley checkpoint on civitai. looks good for sayuri. probably because her realistic design. I also found the doaxvv lora is making the background grainy when upscaling. So I tried to go as lowest as possible. That is 0.4 strength. Going lower the face is not the same as game. Turning lora off works too if going with low denoise. You can always inpaint or adetailer later. Edited February 8 by huehuehue9001 1
5250 Posted February 8 Posted February 8 6 minutes ago, huehuehue9001 said: I think the doaxvv lora is making the background grainy. try lora at 0.4 strength. Oh! Thank you for your advice. I will try!
yoshi09 Posted February 8 Posted February 8 (edited) みさき、たまき、ほのか、エリーゼ、フィオナ Edited February 8 by yoshi09 10
huehuehue9001 Posted February 11 Posted February 11 (edited) valentine + birthday. Spoiler Can't forget the secret ingredient for Owner-sama. Full of love. Edited February 11 by huehuehue9001 5
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now