⚡ EARLY ACCESS — Limited time only. This model is available for direct download for 15 days only. After that: Generation Only on Civitai. No more downloads here. Get the best pricing in the official store.
→ Get it now for much cheaper prices, skip the wait:
🛒 Ko-fi · TensorArt ✅ Card & PayPal accepted — no crypto, no Buzz, no queues ✅ Instant download · Priority support
Also available:
Event Horizon ZIT 2.0 (Kofi)
Event Horizon XL 7.0 (Kofi)
Get the complete Event Horizon collection with all the future updates included→ Ko-fi
⚙️ Recommended settings with Z-Image Turbo versions:
8 steps, cfg1, Euler or DPM++ 2s a RF sampler, Normal or Simple or Beta scheduler, 832x1216, 1024x1536 or 1280x1920 pixels
Upscaling:
8 steps, cfg1, Euler sampler, Normal or Simple or Beta scheduler, 1.5x upscale
Important: remember that you need to place qwen3_4b.safetensors in the text_encoder sub folder and ae.safetensors in the VAE sub folder for Z-Image Turbo to work properly.
⚙️ Recommended settings with SDXL versions with DMD2 lora:
Sampler: LCM Karras / Exponential
CFG: 1-1.3 (recommended is 1)
Steps: 8 - 12 steps
Resolution: 896x1152 / 832x1216
Clip Skip: 2
VAE is bundled in, you don't need to download anything else.
⚙️ Normal SDXL Recommended Settings:
Sampling method: DPM++ 2M SDE Karras
VAE: is already Baked In
Sampling steps: 30+ Steps
Resolution: 832x1216 / 896x1152 CFG: 3-5
Clip Skip: 2
Upscale: 1.5 upscale (0.3 - 0.35 denoise)
Upscalers: 4x_NickelbackFS_72000_G / 4x_foolhardy_Remacri / 4x_NMKD-Siax_200k
Description
Merry Christmas everyone!
V. 4.0: Improves over 3.0 in lighting and creativity, slight improvement on details.
V 4.5: Biggest leap in IQ so far into the EH series (Bigger than 2.5 to 3.0). Better at artstyles, better textures, more details, creativity and much more refined image quality (available now on TensorArt).
FAQ
Comments (13)
Interesting to see, that you went from realism to Pony style, not otherwise.
This model is so much fun to use.I tried all of your showcase prompts with same seed, 1 cfg, 12 steps, lcm/karras, dmd lora, none of them producing same outputs as showcase, all different, sometimes they even looks like different checkpoint,
Are you using Comfy? and please notice that in some samples, (like prompts that start with "eh" are embeddings of my own that you can download here (professional artistics/ professional fantasy and such... a lot of personal aesthetics I really dig) and are part of the eco system of EH. All the embeddings are on civitai to download for free. Check suggested resources out if you like. You'll need those if you wanna replicate the outputs.
But yeah, everything here is made only with EH 4.0. No refiners were used, no extra checkpoints. Just some of my own embeddings i've been working these months and I used it in some samples.
Cheers!
@ElectricDreams ty so much for your reply, now i understand (im not using embeddings). your models is fun to use. marry christmas
Prompt adherence is terrible for 4 and 4.5. Can't believe I spent money on this. Most of the time it just ignores what you enter. Negatives can't be used without ruining the image completely. Not using dmd2 ruins every single image.
Oh sorry to hear that! Prompt adherence has been always the weakest part of SDXL overall. I'd try to improve that in the future. I always use dmd2 lora, because it always improves the outputs tremendously, not only in the checkpoint but in every other checkpoint, so i don't know, but i guess it depends on the scheduler and other settings. Results may vary a lot. What configuration are you using? thanks for commenting!
Prompt adherence is way better, than in the early versions, but its not realistic anymore.
@ElectricDreams Using forge trying to edit some of your example pics. Just changing the clothes is difficult. It ignores most poses. Avoids nudity across the board. Never had these issues with other XL models. Even version 3 generally does what you tell it, but 4 and 4.5 hardly obey prompts at all.
@bougyakumahou I'll look into it. Thanks for commenting.
@ElectricDreams just try something simple like, women wearing an untied kimono, bare stomach, bare chest, crossed arms. all you get is a woman in a kimono. the rest is ignored.
@jarrettjermon678 levels of photorealism can vary between versions. Normally a photorealistic checkpoint is not very good at artstyles or overall creativity (which is the absolute priority in this checkpoint). At least that happens in the SDXL realm (i don't know about Zit, Flux or other models). So while 4.0 is slightly less photorealistic than 3.0 is a lot better at artstyles, aesthetics and other features. In any case, that's why i made Event Horizon Nexus which focuses in photorealism, so i didn't have to compromise creativity in Event Horizon XL. So people looking for photorealism might want to try that one instead. Thanks for your comments!
@ElectricDreams the realism is very good, but if it won't listen to your prompting it isn't very useful.
@bougyakumahou the new version will improve prompt adherence vastly. Be at ease. I already have an operative prototype that is quite a leap on that department. It just gonna take a little bit more of time. In any case, thanks for your insights.