I’ve been playing around with various mixes of Stable Diffusion Checkpoints, Loras, Sampling Methods, and some of the results are almost of Mid Journey quality. Just need to work out how to sharpen and upscale (enlarge) the small resolution SD images with SD’s controls (or Photoshop)…Any text-to-image prompt setup above 800x800 gets strange / deformed results…
1 Like
Nice! Would you mind sharing some prompts?
1 Like
Here are some of the prompts I can find. The Checkpoints and Loras were acquired from the Civit Ai website…The negative prompts were copy-pasted straight from the website’s image examples too…
^Freyja on her bed
Positive Prompt: photograph of a teenage girl ((with shining emerald green eyes)), Israeli girl, 14 y. o. , smiling, long black hair, dense muscles, fitness model, tanned olive skin, lying on a double bed, dressed in a oversize military tracksuit (oversize jumper), photo - realistic, luxury villa bedroom scene, 8K, shot on a Leica DSLR, f/ 2. 8, intricate detail, bedtime lighting,
Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render, (tiara), (crown), (bad hands)
Setup: Steps: 140, Sampler: DPM++ 2M SDE Karras, CFG scale: 10, Seed: -1, Face restoration: CodeFormer, Size: 700x700, CkPt: cyberrealistic_v31, Version: v1.3.2
^As above, except with the ‘deliberate_v2’ Checkpoint…Should have been an Australian army tracksuit, but I forgot that part…
^Audi at a villa prompt (lotto ticket not included)
Positive Prompt: Green Audi S5 parked on the driveway of a European villa, car photography, photo realistic, ultra - detailed, shot on a DSLR, 4K
Negative Prompt / Setup: See above. I recommend the ‘hypercars’ Lora for a customised look.
Setup: Steps: 120, Sampler: DPM++ 2M SDE, CFG scale: 7, Seed: -1, Face restoration: CodeFormer, Size: 700x700, Checkpoint: cyberrealistic_v30
^Krista visiting her father's homeland
Positive Prompts: full body photograph of a adolescent German girl, 14 y.o., youthful complexion, blonde hair, long ponytail, smiling, blue eyes, dressed in a red hiking coat and jeans, Germany, Blackforest scene, 85mm lens, shot on a Pentax DSLR, 8K, f/2.8, 48 megapixels, HDR, intricate details, photo-realistic, shot on a Pentax DSLR, 48 mega pixels, 85mm lens, f/2.8, sharp focus, 8K resolution, detailed skin, (masterpiece:1.2), (best quality)
Negative Prompts: Negative prompt: Negative prompt: canvas frame, cartoon, 3d, ((disfigured)), ((bad art)), ((deformed)),((extra limbs)),((close up)),((b&w)), blurry, (((duplicate))), ((morbid)), ((mutilated)), [out of frame], extra fingers, mutated hands, ((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))), out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), Photoshop, video game, ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, mutation, mutated, extra limbs, extra legs, extra arms, disfigured, deformed, cross-eye, body out of frame, blurry, bad art, bad anatomy, 3d render, (tiara), (crown), (bad hands)
Setup: Steps: 100, Sampler: DPM++ SDE Karras, CFG scale: 17, Seed: -1, Face restoration: CodeFormer, Size: 600x800, Checkpoint: cyberrealistic_v30, Version: v1.3.0
^Hannah's Alpina B6 (second lotto ticket required)
Positive Prompt: masterwork photograph, Alpina B6, 2016 model, metallic blue paint wrap, tuner car body kit, chateau driveway scene, 8K, intricate details, shot on a Sigma DSLR, natural daylight lighting, marketing photography, professional photography, luxury driveway setting lora:hypercars:.5
Negative Prompt: As above
Setup: Steps: 120, Sampler: DPM++ 2M SDE Karras, CFG scale: 12, Seed: -1, Face restoration: CodeFormer, Size: 800x600, Checkpoint: I Can’t Believe Its Not Photography, Script Option: Sonar sampler (default setting)
^Naida's military days
Positive Prompts: photograph of a Siberian woman with ginger hair and ((freckles:1)), strong, muscular, dressed in a winter military uniform, professional photography, 48 mega pixels, 85mm lens, sharp focus, 8K resolution, detailed skin, (skin pores), (masterpiece:1.2), (photo-realistic:1.2), (best quality), (detailed skin:1.3), (intricate details), winter forest setting
Negative Prompts: The usual
Setup: Steps: 80, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: -1, Face restoration: CodeFormer, Size: 600x800, Checkpoint: pirsus Epic Realism, Script Options: Sonar sampler: (default settings)
^Lora update experiment (forgot to install the update first...Jumper dress: puffer jacket...)
Prompts: full body photograph of a beautiful German girl wearing a red puffer jacket, (very long blonde ponytail), (teenager), 48 mega pixels, 85mm lens, sharp focus, 8K resolution, detailed skin, (masterpiece:1.2), (photo-realistic:1.2), (best quality), (detailed skin:1.3), (intricate details) lora:jumper_dress:.5
Setup: Steps: 140, Sampler: DPM++ 2M SDE Karras, CFG scale: 10, Seed: -1, Face restoration: CodeFormer, Size: 600x800, Checkpoint: artErosAerosATribute_aerosNovae, Script Options: Sonar sampler: (default settings).
^An Aussie walks among us
Prompts: masterwork photograph, Holden HSV, metallic green paint wrap, tuner car body kit, Australian car show scene, 8K, intricate details, shot on a Sigma DSLR, natural daylight lighting, marketing photography, professional photography, car salon setting lora:hypercars:.5
Setup: Steps: 120, Sampler: DPM++ 2M SDE, CFG scale: 12, Seed: -1, Face restoration: CodeFormer, Size: 700x700, Checkpoint: UmiAIAphroditeV2_v20
Oh, honourable mentions to the following (most reliable) Checkpoints:
Cyber Realistic v31
Pirsus Epic Realism - Hyper Detailed
Dream Like Photo Real
Epic Realism: New Era / New Century
Reliberate
PS.
We need the '80s back.
1 Like
Thanks so much. I’ll see if I have any luck with these
Also, where are you using it that you can write such long prompts? Is it your local SD version or somewhere else?
1 Like
I have a local installed copy of SD Automatic 1111. Here’s a useful guide.
1 Like
I got some squids for @DomiSotto.
Summary
Latest prompt:
drawing of a purple squid alien stacked on top of a squid, indigo and muscular monster, hyperrealistic cartoon, vivid colors, splash art, unreal engine
Most of the above were done without the stacked part though.
We could try to manually stack them up and then evolve them. Let me know if you like any of the above.
1 Like
Well, it looks like Dream went to only theough the app, which is 10x more anoying to get pucs off of, although just as free to use. Oy. Think its a move to start getti g more paying members, now that aome of the filters are qorking right.
1 Like
Omg, 3, 4 and 6 are equally awesome. Like, that’s exactly what I look for style and complete wackiness. Gosh, now I want to totally get the lite novel illustrated via AI. Just gorgeous. Ah… I know you are busy, but can you teach me or co-author the story as artist?
1 Like
Yes, they did. At least on the phone. Haven’t tried on my laptop since the change. Annoying but as long as they keep the free plan, it’s still worth it.
I honestly wouldn’t know how to teach you. You just have to start and experiment until you find what works and that depends on individual prompt and platform. And then you need to be patient. Sometimes you get lucky and you get what you want in seconds. Other times, it might take weeks.
We could try to collaborate if you’d like. Not sure how it will go but until we try, we won’t know.
1 Like
The way I write the story I have the mini description of each character (there is 7 aliens (6 left) that are in the verse). I can try using the software you suggested for generating fantasy versions of my locations if I can upload pictures in it, but the strange aliens like that would probably be harder. I can try Wombo Dream for them and if I can’t get anywhere with one of the images, I can ask you to give it a shot? I am happy to add you as a co-author and give an enthusiastic credit.
Sure. Give it a try and let me know if you need help.
All that does is slow down my chronic generations. Lol
For an experiment I acquired the SD lora detail tweaker and tested several random Mid Journey prompts from my gallery. Copy-pasting, no edits. Some of the results are better than expected. Still receive postage-stamp sized images though. I also learnt how to use a VAE, if that makes any difference.
PS. Still waiting for a decent SD up-scaler, or a local installation option for (uncensored) Mid Journey.
1 Like
I love the style you gave them.
It ties it all together, making it uniform.
Also, I love the silliness of the book. It’s just fun.
1 Like
Crap. These are great. Makes me regret that I didn’t invest more time into setting up my own yet. I’d need hardware though.
A while back I read somewhere that it’s pretty cheap to use AWS for it, but I don’t know how much to trust this opinion. I’d feel better with my own hardware too. Sigh.
1 Like
That is ridiculously cute.
1 Like
Alright. had to read up on how I want to store backgroundless photos, and that means a png file and not a gif. They all compress somewhat, but the older ones aren’t as friendly. (For example, I used to get horrible interlacing/speckling from 3rd-hand jpg files back in the day, and I don’t think that ever wholly cleared up (and is why I use my GIMP files until done, as much as possible.)
But anyway, starting to cut out my “parts files”…
Twice the size of original to keep the cuts from being too blocky, so best to use it at half it’s size or on something equally enlarged. shrugs
1 Like
I’ve heard the next challenge for AI development is to reduce the system requirements for local installations. Not sure if graphic / art AI software is included, but it would be nice. My 12GB 3060 becomes a room heater during most SD experiment sessions.
A recently discovered issue with SD is when a Checkpoint or Lora is updated by the SD web browser, the old version is not over-written / deleted. I now have several listing for the same Checkpoint. A tad difficult remembering which ones conjure the best images (depending on the prompt subjects).
Some more SD prompt experiments…
Krista enjoying her birthday gift from her Aussie bodyguard, and close friend, Dov. A tailored Kangaroo leather jacket.
^A before and after photo.
Once again I try to prompt for Zorik's character portraits only to receive great images for his son, Leon.
^let’s assume that’s a micro Uzi. Leon takes library etiquette quite seriously…probably at his step-sisters insistence…(Bonus points for trigger discipline).
MJ still has SD beat for the quality of high resolution images. Or I need to do more research into SD’s tool sets.
2 Likes