usamaehsan / ltx-voice-one
LTX voice: text to speech with voice design or a saved voice_id, one warm pool (merged ltx-voice + lock).
usamaehsan / longcat-avatar
Talking avatar: one portrait photo + speech audio -> lip-synced talking video (LongCat-Video-Avatar 1.6, MIT). 8-step distilled, 25 fps, audio muxed in.
usamaehsan / klein-4b-lora-lab
Research only: train and test a LoRA on FLUX.2 klein 4B (Apache 2.0) edit pairs in one run.
usamaehsan / hunyuan3d-fast
Test: Hunyuan3D-2 with baked weights (cold-start)
usamaehsan / flux2-klein-4b-edit-fastboot
Test only: flux2-klein-4b-edit with a slimmer setup (cold-start test). Use usamaehsan/flux2-klein-4b-edit.
usamaehsan / whisper-ears
Whisper turbo transcript + audio X-ray in one pass: speaker turns, tone per line, laughs/coughs/sighs with times, 5-clip custom sound detectors, and an LLM-ready report.
usamaehsan / llm-ears
Research: gives a frozen LLM (Qwen3-8B) ears through a closed-form linear bridge from a frozen Whisper large-v3 encoder. No gradient training. Experiment mode measures WER on LibriSpeech.
usamaehsan / ltx-voice-speed-test
Speed/cost test copy of usamaehsan/ltx-voice (designed voices from text, no audio input). Not for production.
usamaehsan / ltx-voice-cheap-test
TEST ONLY: cost experiments for the LTX-2.5 voice model (batching, vocoder precision). Not for any site.
usamaehsan / ltx-voice-lock-test
Test: LTX-2.5 voice with fixed library voices (voice_id). Designed voices made by the model itself, no audio input, no cloning.
usamaehsan / ltx-av
LTX-2.5 22B distilled: fast expressive text-to-speech (default) plus optional text/image/audio-to-video with native audio, in one model. Designed voices from text, no cloning.
usamaehsan / ltx-voice
LTX-2.5 voice only: expressive text-to-speech from the audio half of LTX-2.5 (no video). Designed voices from a text prompt, no voice cloning.
usamaehsan / qwen3-tts-voice-design-test
TEST ONLY - do not use. Throwaway test of a faster build of usamaehsan/qwen3-tts-voice-design.
usamaehsan / qwen3-tts-voice-design
Text-to-speech in a new voice designed from a text description (10 languages). Qwen3-TTS VoiceDesign, Apache-2.0. No audio input, no cloning; descriptions that name a real person are refused.
usamaehsan / chatterbox-tts-v18
Fast batched multilingual text-to-speech (23 languages) on Chatterbox Multilingual. Preset voice library, no voice cloning, no watermark, max_seconds cap that stops at a sentence end.
usamaehsan / chatterbox-mtl-v18-test
TEST ONLY - do not use. Throwaway test of chatterbox-tts-v18.
usamaehsan / supertonic-3-tts
Fast multilingual text-to-speech (31 languages, 10 preset voices, no cloning). Supertonic 3 weights, OpenRAIL-M: outputs must be disclosed as machine generated; see the README for the use restrictions.
usamaehsan / ace-step-bgmusic
Cheap fast instrumental background music (ACE-Step 1.5 turbo, MIT)
usamaehsan / chatterbox-mtl-batched
Chatterbox Multilingual V3 (MIT) with batched sentence synthesis: a whole story is decoded in one pass instead of chunk by chunk. Voice cloning from a short reference, 23 languages.
usamaehsan / qwen-image-edit-2511-nunchaku
Qwen-Image-Edit-2511 Lightning, Nunchaku SVDQ INT4 (official nunchaku 1.2.1 + 2511 shim)
usamaehsan / qwen-image-edit-2511-lab
Qwen-Image-Edit-2511 cost lab: FP8 + torch.compile + SageAttention, tunable steps/reference size
usamaehsan / qwen-image-edit-2511-fast
Qwen-Image-Edit-2511 with the 4-step Lightning LoRA fused offline, FP8 on L40S.
usamaehsan / ernie-image-turbo-fast-h100
ERNIE-Image-Turbo tuned for cost: FP8 + torch.compile + lean forward, 1-8 steps, optional hires refine
usamaehsan / ernie-image-turbo-fast
ERNIE-Image-Turbo tuned for cost: FP8 + torch.compile + lean forward, 1-8 steps, optional hires refine
usamaehsan / z-image-turbo
Z-Image-Turbo text-to-image, 8 steps
usamaehsan / ernie-image-turbo
ERNIE-Image-Turbo text-to-image, 8 steps, strong text rendering
usamaehsan / llada-image-turbo
LLaDA-Image-Turbo text-to-image (fp8), strong text rendering
usamaehsan / mage-flow-edit-turbo
Mage-Flow-Edit Turbo (4B, MIT) - instruction-based image editing in 4 steps.
usamaehsan / boogu-image-0.1-edit-turbo
image edit bench
usamaehsan / firered-image-edit-1.1
image edit bench
usamaehsan / flux-kontext-fastest
FLUX.1 Kontext dev, int4 + turbo 8-step. Cheapest quality image edits on L40S.
usamaehsan / ltx-2.5
usamaehsan / ltx-25-distilled
usamaehsan / flux2-klein-4b-edit-fp8
FLUX.2 klein 4B image edit, fp8 quantized for max speed on L40S
usamaehsan / flux2-klein-4b-edit
usamaehsan / qwen-image-edit-fastest-2
400+ images in 1$ 1024 resolution// 2s /img on L40S GPU, check description and examples
usamaehsan / qwen-image-edit-fastest
300+ images /1$ approximately - check examples. Quality is good even at 2 steps
usamaehsan / qwen-image-4bit
good for simple tasks - running on l40s
usamaehsan / flux-multicontrol-inpaint
usamaehsan / voices
usamaehsan / voice
usamaehsan / flux-controlnet
usamaehsan / flux-multi-controlnet
Fast FLUX DEV -> Flux Controlnet Canny, Controlnet Depth , Controlnet Line Art, Controlnet Upscaler - You can use just one controlnet or All - LORAs: HyperFlex LoRA , Add Details LoRA , Realism LoRA
usamaehsan / flux-multicontrolnet
multi controlnet union pro <-
usamaehsan / juggernaut-xl-x-adapters-x-lightning
usamaehsan / instant-id-x-playground-v2.5-xl-lightning
usamaehsan / multi-controlnet-x-ip-adapter-vision-v2
usamaehsan / swap-sd
Experimental & for non-commercial use only
usamaehsan / instant-id-x-juggernaut
usamaehsan / instant-id-x-yamermix-v8
usamaehsan / instant-id-x-copax-timeless-xl
multi Instant ID + Pose
usamaehsan / instant-id-1
usamaehsan / instant-id
fast instant-id
usamaehsan / instant-id-x-better
usamaehsan / controlnet-x-realistic-vision
usamaehsan / controlnet-x-ip-adapter-realistic-vision-v5
Inpainting || multi-controlnet || single-controlnet || ip-adapter || ip adapter face || ip adapter plus || No ip adapter
usamaehsan / controlnet-x-majic-mix-realistic-x-ip-adapter
works with inpainting and multi-controlnet + single-controlnet || ip-adapter + without ip adapter
usamaehsan / controlnet-lineart-brightness-tile-inpainting
controlnet-lineart-brightness-tile-inpainting + low res fix with tile
usamaehsan / lineart-x-brightness-x-tile-x-inpainting-x-realistic-vision-v5
usamaehsan / multi-controlnet-x-consistency-decoder-x-realestic-vision-v5
Multi-Controlnet + consistency-decoder + INPAINTING + realestic-vision-v5 + Prompt-Weight + Single-Controlnet
usamaehsan / controlnet-lineart-inpainting-x-realistic-vision-v5
usamaehsan / realistic-vision-v5-inpainting
Better than SDXL Inpainting
usamaehsan / controlnet-1.1-x-realistic-vision-v2.0
controlnet 1.1 lineart x realistic-vision-v2.0 (updated to v5)
usamaehsan / dual-style-gan