<h1 id="artsy-dream-pkml0buwu">ARTSY DREAM</h1>This Project has concluded with the sixth version.<h3 id="check-out-my-newest-model-in-the-artsy-series-artsy-vibe!-ghkigt9uk">Check out my newest model in the Artsy Series, <a rel="ugc" href="https://civarchive.com/models/1162948">ARTSY VIBE!</a></h3><h3 id="version-6-9bzk2adkt">Version 6</h3><pre><code>(release notes - 12/28/2024)

Okay so I lied. I'm an addict who simply cannot stop tinkering with this model.

Fortunately, that just means I have a more improved version ready for the world. Version 6 has increased fidelity with more careful balancing between compositional creativity, prompt adherence, and meaningful fine details.

Whether this version will be the final version, I cannot say. But this one does feel more complete than the last.

Happy New Year!

Files still uploading - FP16</code></pre><h3 id="version-5-60ieztexf">Version 5</h3><pre><code>(release notes - 12/06/2024)

After a few months of working on the Artsy Dream project, I've spent time curating my own data sets in order to achieve the closest possible representation of my personal aesthetic tastes... And I think I've gotten there, at least to the extent that Flux Dev can take me.

This version of Artsy Dream has some compositional and textural improvements, leaning more into the less-realistic and more fantastical nature of the model. Though, of course, it is still capable of photo-realism.

In some concepts/styles, this model is only a minor improvement/change compared to v4. In others, there are substantial stylistic changes.

Enjoy Version 5!</code></pre><h3 id="version-4-5bonn7c5c">Version 4</h3><pre><code>(release notes - 11/17/2024)
After much indecision about how I was going to release this model, I have finally decided that it will be the fourth installment of my Flagship Model Merge. 

This model marks a notable pivot in aesthetic direction. However, it remains part of previously named "Artsy Flux", and now rebranded "Artsy Dream" model page. However different, this model still utilizes much of the same dataset, producing comparable outputs.

This version demonstrates a substantial leap forward in fidelity, style versatility, and prompt adherence. More complex and precise merging methods of very specifically chosen block layers have resulted in a far more polished model. Enjoy greatly expanded 2D illustrative capabilities as well as the ability to create more abstract compositions. 

Unleash your creativity with the new "Artsy Dream" Version 4.
Enjoy!</code></pre><h3 id="version-3-f8kpytiqw">Version 3</h3><pre><code>(release notes - 11/3/2024)
Building on the improvements in version 2. I spent my time aiming to tame the noise/artifacting that inherently happens when trying to invoke the level of detail that this model has. The challenge was to retain the compositional creativity and detail, and in that, I have succeeded. 

Anime and 2D imagery in general will be possible with this version, but definitely not with the stylistic versatility as version 2 (and may take more coaxing with prompts). Based on the images that people have posted to the model galleries, I feel like the focus should be creativity and versatility in styles most commonly used. It can still do various 2D and anime styles, but functionality may not be as thorough as version 2. Everything else has been vastly improved. (there will be examples provided for reference)

(lower LoRA strengths if you're having compatibility issues)</code></pre><h3 id="version-2-1stq41p35">Version 2</h3><pre><code>(release notes - 10/28/2024)
The newest version of Artsy Flux improves on the advantages of version one and attempts to fix issues the original version had. Random messy noise has been greatly reduced in most contexts while maintaining the level of detail and creativity that the model is known for. Color grading has been enhanced substantially by training in properly color graded data, reducing outputs that appear to have a photo filter.

The capability to create anime in various styles has been expanded on. The model is now capable of 2D, 2.5D, and 3D anime.

Pretty happy with this version, so it may be some time until there is another. Enjoy!

(lower LoRA strengths if you're having compatibility issues)</code></pre><h2 id="model-description-4pye8zxcs">Model Description</h2>This Model is a sophisticated, carefully crafted merge of various self-trained LoRAs designed into introduce compositional and detail concepts into Flux.1 Dev in order to invoke the best aesthetics and qualities that Flux has to offer. It excels at realism with diverse environments and stylistic compositions, but is also capable of 2D imagery, adding a comparable amount of detail and artistic flair. Images produced have more cohesive color grading and lighting than most base Flux.1 Dev outputs. I want to emphasize that this model was created for the main purpose of creating images with my personally preferred aesthetics. Future versions may include more or remove style based on user usage or my preference. I will leave past versions up to account for anything that may be removed in newer versions. This artfully curated merge allows you to get the best out of Flux WITHOUT the use of detailer LoRAs.Achieve compositions and levels of detail that can even rival Flux Pro at the fast speeds of any version that works best for your local configuration:Generations taking far too long? Try my Schnell model <a target="_blank" rel="ugc" href="https://civarchive.com/models/920261/artsy-lite">ARTSY LITE</a>. (4 steps)<h3 id="versions-available:-t1pvkblhc"> VERSIONS AVAILABLE:</h3>FP16 (.safetensors) FP8 (.safetensors) Q8_0 (.gguf) Q4_K_S (.gguf) Q3_K_S (.gguf) (v3 and up only) VERSIONS THAT WILL NEVER BE AVAILABLE (Do not ask): NF4 AIO (clip_l, t5xxl and vae included) SOME GGUF QUANTS MAY NOT BE FULLY FUNCTIONING IN FORGE UI DUE TO COMPATIBILITY ISSUES CURRENTLY BEING ADDRESSED BY ILLYASVIEL. These models use the same <a target="_blank" rel="ugc" href="https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/clip_l.safetensors">clip_l</a>, t5xxl (<a target="_blank" rel="ugc" href="https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/t5xxl_fp16.safetensors">FP16</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/t5xxl_fp8_e4m3fn.safetensors">FP8</a>), and <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.safetensors">ae.safetensors</a> as Base Flux.1-Dev &amp; Flux.1 Schnell. (Images attached to each version have been generated with their respective model sizes/quants for accuracy and honesty. Though you'll be surprised at the level of detail achievable with this Q4_K_S.) REGARDING NSFW: I do not add any NSFW data to my models. It can produce NSFW material in about the same capacity as Flux Dev, and I will not be adding any NSFW capabilities to any future versions. ENJOY!

Artsy Dream

artsyDream_v4Q4KS.gguf

v4 - Q4_K_S

v6 - FP16

v6 - Q3_K_S

v6 - Q4_K_S

v6 - Q8_0

v6 - FP8

v5 - FP16

v5 - Q3_K_S

v5 - Q4_K_S

v5 - Q8_0

v5 - FP8

v4 - FP16

v4 - Q3_K_S

v4 - Q8_0

v4 - FP8

v3 - Q3_K_S

v3 - FP16

v3 - Q4_K_S

v3 - Q8_0

v3 - FP8

v2 - FP16

v2 - FP8

v2 - Q4_K_S

v2 - Q8_0

v1 - Q8_0

v1 - Q4_K_S

v1 - FP8

v1 - FP16

Files with SHA256: 0835c2a543416daee41d5798b90df035f586b6b8241f61fe30cfe8097140e160