Consistency V3 Checkpoint [FLUX.1D]

This is an old model. Use the newest Consistency Simulacrum.

It does everything this does; https://civitai.com/models/477291/consistency-v3-flux-d1

This is a merge of Flux1 Dev with Consistency-V3-Flux-D1. It legitimately doesn't need the clips because the clips weren't trained by the lora.

Strength; 0.8 << might be a little high

Convenience:

For those just getting started, I've made a convenient pair of merges that takes less than about 16 gigs of vram at times, but it's also between 16 and 20 gigs at other times.

FP8 T5+vae has the t5 encoder embedded into it. When merged together it runs faster and more effectively in COMFYUI but I'm unaware of how well it runs in forge currently.

Advanced:

For those interested only in the style, concept, and individual-centric pose control in conjunction with your own system or systems; go grab just the lora.

FP8 noclips has no built in clips, so if you plan to download or run your own custom clips be sure to take this route.

Other Notes:

• Loras I've been testing that usually work with the separate system are showing a lack of context awareness, so there's definitely a time and a place for high powered individual consistency.

• The full merge loads faster, generates faster, has less vram requirement, and still retains the core details and information. It also took forever to upload, so please enjoy.

• I'll be taking another look at GGUF today. I had a lot of trouble implementing it when I tried, as the subsystems and the core systems aren't exactly in my forte of skillset.

• GGUF isn't an easy system to simply just run currently. There's a lot of edge cases and problems apparently, which is why the current systems are defiant of it's pure utility. It'll take some time but I plan to implement a Q2, Q4_0, Q4_K_M, Q6, and Q8 implementation to ensure that EVERYONE can play with versions of Consistency.