🌀 Abstract Surrealism: A Fusion of FFusion.AI & DALL·E 3 Creativity 🎨
Diving deep into the intricate world of abstract surrealism, we've merged the visionary prowess of FFusion.AI with the distinct style of DALL·E 3. This artistic experimentation is bound to push boundaries and redefine perceptions.
Hardware Setup:
GPU Configuration:
8x A6000 cards
Loras:
LoHA Basic:
Alpha:
1.0Dim:
32Conv Dim:
32Conv Alpha:
1Use CP:
FalseAlgo:
loha
Lite Weight LoRA
Highlight: Elevated
network_alphafor sharper insights.
Heavy LoRA
Dim:
128
Lora FA
A classic integration, always evolving.
Locon with Experimental VIT Prompts
Exploring uncharted waters with VIT-based prompts.
Locon - High Alpha Version
An enhanced alpha ensures richer details and contrasts.
LoHA FULL Algorithm:
Alpha:
0.5Dim:
64Preset:
fullAlgo:
lohaDropout:
0.0Authored by:
FFusionAI
🌐 FFusion.ai Contact Information
Proudly maintained by Source Code Bulgaria Ltd & Black Swan Technologies.
📧 For collaborations, inquiries, or support: [email protected]
🌍 Locations: Sofia | Istanbul | London
Connect with Us:
Our Websites:
Description
LoHA FULL preset:
Alpha:
0.5Dim:
64Preset:
fullAlgo:
lohaDropout:
0.0
"ss_network_alpha": "0.5",
"modelspec.prediction_type": "epsilon",
"modelspec.tags": "FFusion.AI, FFai.eu, FFusion, Lora, Lycoris, Lokr",
"modelspec.timestep_range": "50,950",
"modelspec.architecture": "stable-diffusion-xl-v1-base/lora",
"modelspec.sai_model_spec": "1.0.0",
"modelspec.author": "FFusion AI",
"ss_network_module": "lycoris.kohya",
"modelspec.date": "2023-10-11T13:40:56",
"sshs_legacy_hash": "ac62302e",
"sshs_model_hash": "0f2d6ef2e1ed0c59250695679285e3cc504326661cc198a6f34d43ce4bd63517",
"modelspec.implementation": "https://github.com/Stability-AI/generative-models",
"modelspec.title": "Dalle3mixLoha",
"ss_v2": "False",
"ss_network_args": {
"preset": "full",
"algo": "loha",
"dropout": 0.0
},
"ssmd_display_name": "DalE-3-FFusion-LOHA",
"ssmd_author": "FFusionAI",
"ss_network_dim": "64",FAQ
Comments (4)
this is my kind of crazy, all pictures look great.
Any advice for prompting would be appreciated, looks like keyword soup.
Another fan of surreal art i see ? :)
The Lora you're seeing is a preview of our coming surreal XL model, training over over a ~month now.... but the text encoder is the cause .and the prompting ...
I understand why the prompts might look like keyword soup... As they are exactly that :) I intentionally mashed them up in a [A] [B] [C] batch sequence, with over 100+ each prompts randomly pushed together in a batch to demonstrate see the Lycoris capabilities compared to the model it self, since I was on a tight deadline to finish the official model's one-month training.
The upcoming model is incredibly flexible and specifically designed to handle surreal requests while NOT mixing some... like the soup you mentioned...
and speaking of soup do check poor bambi with a bad prompt and a loosy encoder = https://ibb.co/HFBqypC
:)
Once it's released, the surreal FFusion model will be the foundation for adding additional surreal Loras to enhance the logic in certain complex scenarios.
As for Lycoris, try combining two or more prompts that you would typically use separately. For example, if you have prompts to generate two different objects, combine them and maybe add some effects. You'll be surprised at the results! But again, the results will also heavily depend on the base current model.
THe loha is trained on FULL not just attn-mlp. But do try mixing like the X and Y mix in Style Pile mod is doing (slowly mixing almost word by word from prompt Y to X to transition the scene. https://github.com/some9000/StylePile#tips-and-tricks
Also do mix several of those lycoris version in the prompt to see how the outcome changes as they are trained differently each.
Remember, the DALL·E add-on and the current Lora are still limited in handling complex mixtures, as the full Lycoris is around 5+GB size and no point in uploading all those. If you're keen on exploring the surreal potential of the XL version, drop me a message. I'd love to onboard you as a pre-release tester to assist fixing any "bambi issues" we missed :)
@idle Hi, thanks for the detailed reply, I kind of worked out the [A] [B] [C] based on the prompt examples, so far I've had mixed results but mostly Bambis. It wasn't clear if A B C are supposed to work together or are their own things whilst contributing to the end result.
I like testing things so I can help in a non committal fashion.
I love this series.
followed & tipped (i dnt knw if i should mention it or not, i generally dnt reveal tips but in this case i m overwhelmed)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.