Previous Thread: >>107427740>Links:>DALL-E 3https://www.bing.com/images/createhttps://designer.microsoft.com/image-creator>4ohttps://chatgpt.com/https://sora.chatgpt.comhttps://copilot.microsoft.com/>Imagen 4 and Nano Banana (Pro)https://gemini.google.com/apphttps://labs.google/fx/tools/image-fxhttps://labs.google/fx/tools/whiskhttps://aistudio.google.com/prompts/new_image>Midjourney [PAID]https://www.midjourney.com/home>NovelAI [PAID]https://novelai.net/image>Perchancehttps://perchance.org/ai-text-to-image-generator>AI videoshttps://pixverse.ai/https://hailuoai.video/>Instructions and download for BIC EDIT script:https://rentry.org/bicedit>Prompt Generatorhttps://tipseason.com/dalle-prompt-generator>DALL-E 3 image generation guidehttps://files.catbox.moe/o4yihf.pnghttps://files.catbox.moe/320rm1.pnghttps://files.catbox.moe/8gge6o.pnghttps://files.catbox.moe/dcokjf.png>Perchance generator listhttps://pastebin.com/XRGqaW4W>Related Boards>>>/v/dall-e>>>/co/dall-e>>>/trash/aibf>>>/aco/dall>>>/tg/slop
inb4 dog botherer
Dammit forgot to put the meta midjourney in the op
>>107471847>rent free...And gross.
Talking about dem dawgs, one could quite unironically guess how much America is fucked-up just by observing that it's ruled by someone who doesn't like dogs and its secretary of homeland security is someone who literally shot them in cold blood. They're worse than Hitler (who liked dogs, but reportedly hit them).
>>107471794https://duckduckgo.com/?q=gergeti+&ia=images&iax=images
Noticed it just by chance/mistake, somewhat interestingly lmarena does have 4k NBP, but only in "battle mode".Pic unrelated, only bad thing about lmarena is its filters...
>>107471441Just tried to register and it gives me this error (with Opera U.S VPN)
MJ vs Z-Image same prompt
>>107477088Actually, it looks like I can use it without a VPN (I'm in Italy). It just doesn't work in Firefox on my end, I've just logged-in with another mail in Edge.
They seem to have some kind of IP verification/check in place for suspicious activity anyway, I've tried to make another account and it got *immediately* suspended when I received the activation code. Now it wants a fucking photo to "verify" that I'm a human...
https://zforfree.com/Z-Image No login required
Random Astarte tiddieshttps://files.catbox.moe/n28k7i.png
Thread is *almost* deader than our football team today.
>>107481947Yikes, that's a very problematic letter
>>107486173found ya
>>107486173oh for fuck sakes. why praytell is the letter Z somehow a problem to you zoomzooms and your offended things exist mindset
>>107486758ZOV
>>107486860yes yes just more gibberish from a zoomzoom. how about you do something you dont normally do and actually answer the fucking question. this retard zoomzoom mindset you can put on your performance outrage show for updoots and virtual signaling but when asked explain go into gibberish and doing anything but explaining it. its like you guys are just trained to be shitlords and run on shitlord script but have no scripts to you know explain things and just act like a subhuman who doesnt understand words.WHY IS THE LETTER Z OFFENSIVE TO YOU ZOOMZOOM AND IT BETTER BE A LEGIT VALID AND CANON REASON NOT SOME RETARD FAKE UPSET IN YOUR BRAIN! ANSWER THE QUESTION ZOOMZOOM! WHY IS THE LETTER Z SUPOSEDLY OFFENSIVE TO SOME?
>Yaz âµ£ is technically just Z written in another script...Oy vey...
>>107489245sovl
>>107488382I am in love
It looks like OpenAI is preparing yet another model, after all. A less yellow one. I got this by entering battle mode accidentally again. Going from what little one can find around, "hazel-gen-2" is an OpenAI model.https://www.reddit.com/r/OpenAI/comments/1phyva1/are_we_going_to_have_gptimage15/https://www.reddit.com/r/ChatGPT/comments/1phywz8/are_we_going_to_have_gptimage15/https://www.reddit.com/r/OpenAI/comments/1pi2nfc/hidden_image_generation_models/Resolution is the same as 4o/GPT-Image-1 in vertical orientation, 1024x1536.Full resolution gen:https://files.catbox.moe/r8gd8r.png
Same identical prompt generated with "hazel-gen-4". I wonder why there are so many of them and which one(s) will get actually released publicly.
And hazel-gen-2 again, this time it actually made the woman a bit voluptuous, which was specified in the prompt, the breast looks fuller. Boobs aside, it might lean towards the beautification, almost de3-like.
>almost de3-likeI meant de3 "vivid" style, of course. I've managed to get hazel-edit-2 using a reference picture. This time I've also benchmarked its speed, it took around 120 seconds.
>>107490073Have one more.
hazel-gen-2 again. I triggered hazel-gen-4 only once and I really can't seem to trigger hazel-gen-3, I wonder if it's bad luck or lmarena even has it in the first place. Anyway this thing doesn't look bad. Weirdly enough, it doesn't look like people are talking about it on xitter.
Same prompt as above, hazel-gen-4. it's actually a very simple, manually written, one from my old de3 archives:>oil on canvas, impressionistic brushstrokes, minimalist manga painting, soft muted pastel colors, close-up portrait: holding her Remington Model 700, 18 yrs Italian woman (brown wolf ears and tail, short straight brown hair in bob cut, fringe, green eyes, hunting attire) sitting above a fallen tree, a lifeless boar before her. Early morning, light rain.I've removed the artist name that the prompt originally used because I half-suspect this thing won't like the names of living artists like GPT-Image-1, but it's hard to say that for sure while testing it like this.I'm *almost* getting the impression that gen-2 vs gen-4 are a bit like vivid vs natural in de3.
>>107491503>Weirdly enough, it doesn't look like people are talking about it on xitter.Everyone was expecting gpt 5.2 to drop today and that turned out to be fake
>>107490603Not badBetter than seedream-4.5
i don't know why but i started making business cards for imaginary strip clubs
hazel seems to not do toriyama style
>>107493119This is much better, to be honest, I hope it will get released soon enough.>>107493181Yeah, it also seems to work well with simple prompts, unlike GPT-Image-1, which in my opinion generates the blandest stuff ever with bland prompts. But it seems to be more consistent than NBP.The name hazel actually reminds of a certain IRBM... I wonder if it will stick, OAI has never been good with names.
>>107494006Did you have his name in the prompt? I kept getting several errors in a row with Sadamoto's name in the prompt, this is what I got when I replaced it with just a reference to Evangelion (similarly to what I do with Sora/4o).
>>107494388i didn't. when i tried a toriyama style gizmo, just now, i got errors, on lmarena.
>>107494065What is ghost-pepper? it's also good
>replace all wolves in the image to tigersnice
>>107490603>. A less yellow onestill has the piss filter shame
This is you vibe-coding.
>>107494519Thank you. It might work similarly to Sora in terms of filters, but the underlying model is seemingly even more pushed towards avoiding to replicate an *exact* style of living/recently alive artists and defaulting towards its own "anime" style... This was in "in the signature style of Studio Ghibli's "Princess Mononoke" (1997)" (and never mind the legs not being very convincing, the style has nothing to do with it).>>107494986Try to redo the same prompt with gen-2, gen-4 looks more yellow to, between other things.
>>107495140>Try to redo the same prompt with gen-2I only get it once in 10 genstoo slow, esp with the Security Verification on every click
>>107494532Never heard of it before, to be honest. Hazels are popular because of picture related, though.
>>107495226Yeah, it's kind of a pain in that place to test this thing in such a random way. It might veer a bit towards the yellow in some instances, though. For the record, this is the gen-4 version of>>107495312same hybrid anime/Eugene de Blaas genre painting style.
>>107495312I only got it once while editing, supposedly another version of nano banana according to plebbit
>>107495386Doesn't it give a different model for editing? Not totally sure, but I think hazel-edit-2 *might* be the companion, editing, model of hazel-gen-2, but another anon above>>107494778got hazel-edit-6, which is also the one for this "Kramskoification" of a Kirghiz gal, that one was mentioned in the reddit post where that anon mistakenly got a list of the new models, but I have no idea what is its generation equivalent (if any, really).So far on lmarena I've only seen gen-2, gen-4, edit-2 and edit-6. Did anyone see others?
>>107495582they're just going to call it gpt-image-2
>>107495386Never mind >>107495582, I thought I was replying to another anon... I've not seen that model yet and I've generated *quite* a bit of stuff in battle mode today. No idea about it, I'm going to report it if I'll get it.>>107495688I guess it would make sense, albeit I wonder what has happened to that gpt-image-1.5 (lmarena doesn't seem to offer it), but the question is how the model will switch.
>>107495766Try to edit pics, don't think it can gen
>>107495841Yep, I'm going to try that, to be honest I've played much more with txt2img rather than editing... Someone on leddit also mentioned the idea that it's related to image-z, but that... kind of looks like a bold guess, to (unironically) use a kind, respectful and friendly euphemism.https://www.reddit.com/r/StableDiffusion/comments/1pif78h/which_model_is_ghost_pepper_in_artificial/Funnily enough, I got hazel-edit-2 three times in a row with this editing. I liked this one the best.
>>107495958No way, Z-image is utter shit. it's gemini or openai 100%
all gens failing now for some reason
>>107496042Most likely. I still didn't get any ghost-pepper, but here's hazel-edit-2 vs hazel-edit-6, just for comparison. Original photo here:https://pinterest.com/pin/8514686771845778/>>107496261Switch gmail account.
>>107496265HG2 seems like the best one, hope they go with it for release.>Switch gmail account.changed the prompt a lil bit and it went through
bananer pro seems to have had a short reign
'photogrammetry orthophoto / texture atlas of a emperor penguin'slightly disturbing
>>107496333I was usually getting that message after generating too many, not really *so* many, pictures in a row or so I thought. Triggering the model's filter normally resulted only in a single picture getting blocked... and lmarena has its own filters applied to the prompt, which give a different message about their TOS, IIRC.>>107496422Competition is good, either way. Also, OAI shooting themselves on the feet with their policies on referencing names and styles of recent artists is kind of bizarre and self-defeating in the current circumstances.
>>107496422Yeah, it's better than NBP. They're going to leapfrog it like they did with Sora 2>Veo3.>>107496653I think they have a real time filter that updates if you prompt the same thing few times in a row. so has to add a few keywords or change it slightly to bypass it.
>>107496726Just tried it, yeah, you're quite right. Well, switching account does work, if one needs to keep the prompt unchanged, but that's still useful to keep in mind, thank you.Still no luck with that "ghost-pepper" model anyway, I wonder how random (or not) the battle mode actually is. Older models like de3 hardly appear in the results (I got imagen3 once today, I think).On another note:>Error 1015>You are being rate limitedTo reset the router or not, that's the problem.
So, I got "ghost-pepper" while prompting nonsense. It's not only for editing, it was a txt2img prompt (I got kind of pissed off after I got NanoBanana in all flavours while trying an unrelated editing prompt, so I tried something else). It looks weird, I wouldn't say it's either related to Google or OAI's model, it *tries* so hard on the Baroque style (it's supposed to be Artemisia Gentileschi) that it even adopts a bizarre retro anatomy for the pig. Not going to waste a picture for this silly stuff, but for reference:https://files.catbox.moe/x9tqzq.jpegFile format itself (JPEG quality 95, no subsampling) is possibly interesting too, both Gemini and OAI models output PNGs. It's the same settings used by Meta actually, but aesthetically it's completely different (I've tried to generate it after removing all the bodily descriptors that caused Meta AI to blush).
>>107497701But Meta AI is just renamed Midjourney and they never test on lmarena Also V8 hasn't even pretrained yet, they keep delaying it.
>>107497763My suspect is that it's a variation of Flux 2, maybe? It also outputs JPEG 95 without subsampling and it feels a bit like its style, in terms of textures and colors. But results for Flux 2 Pro, using the same prompt, are even sillier:https://files.catbox.moe/tnse8u.jpeg
>>107497808Yeah I don't think it's as good as the hazel-modelsstill new models are always welcome
Man i don't know how anyone uses Bing's creation tool. No matter what I try to make, if it has a woman in it, I get the dog.
what are the least censored cloud models? openai's seem pointless to use with how filtered they are
>>107498291Chinese ones. Seedream for example
>page 11
Anyway I think NBP might be better at emulating (some?) specific painterly styles, this is hazel-gen-2 version of the same prompt I used for thishttps://desuarchive.org/g/thread/107283178/#107296864hazel-gen-2 is better at respecting the angle that I had actually specified and it knows that (Cretan Hound) dog breed, but visually it doesn't remind as much of Bouguereau.
I wouldn't hate young women if they knew their place
hazel doesn't seem to want to generate pokemon, or at least not gengar
For the record, the new OpenAI models are being tested on another website called designarena, too. They use different codenames there, what appears to be hazel-gen-2 is called just Hazelnut, whereas the lookalike of hazel-gen-4 is called Chestnut. Its own filters are even stronger than lmarena's ones, so it's basically useless if you want to generate a woman with even minimally sexualised words in the description... but I guess that might not matter for some.
>>107505128same prompt. nice when i get the combo of nana pro and hazel on lmarena
pig things are happening
>>107503153Nothing beats style reference from MJ. it's a shame no other model has this functionality. I don't think the meta.ai version has it.
>>107507627Yeah, it looks pretty much like a different model when using references, from what little I've tested (it's even more prone to get filtered when using references). It's quite possible that it is a different (older, in-house) model. Well, maybe they'll finish to integrate/improve it sooner or later.
One day AI will understand the concept of trigger discipline.
Getting a bunch rejections today with hazel-2
>>107508783Did the prompts have anything in common? Mentions of artists and/or copyrighted characters? I'm chilling with analogue styles with NBP again this evening, but earlier this afternoon the hazel models seemed to work fine on my end.
>>107508835No, same stuff I did yesterday. either openai cucked the models already or lmarena is acting up.
>>107508885I would lean on the latter if it's not consistent, I get often enough issues with Imagen/Gemini's models on my end. I'm not sure about what you were trying to generate, but I lazily (and luckily) ran this prompt again just now and it worked with hazel-gen-2. It was flux-2-flex which failed to generate anything, instead.
Bump.
New thread:>>107514057>>107514057>>107514057I've added Meta to the list, mentioned that it uses MJ just in case anyone missed the bit of news.