sliders-conceptmod

Shared algorithm and research core for model-specific slider releases. The installable concept-slider-core package now powers Anima Concept Sliders, including native particles, training primitives, ComfyUI and LoRA fitting. See the shared-core architecture and Candlelit / Moonlit samples and weights. YuE2 and Music 3 retain their existing implementations; future integrations can pin the same core after their own compatibility checks.

Learn a musical control, then turn it with a slider. Keep the caption, lyrics and seed fixed while changing voice or genre with a numeric strength.

YuE2's routed-particle method is our preferred formulation. It learns the edit through a paired-error adversarial game, with a regularized particle cloud inside the adapter, drawing on ParticleGAN. This is the cleanest expression of the approach in this fork: adversarial teacher matching without output MSE, feature matching, lyric hold or ending supervision. The math below spells out both the game and its regularizers.

This is a substantially divergent fork of Concept Sliders, with its own music objectives, adapter architectures and listening tools. It also includes MiniMax Music 3, other language-model experiments, modern image/video backends and the inherited diffusion-slider code.

Listen · YuE2 math · Training evidence · Get started · Models · Other formulations · Evaluation · Repository map

Listen and download

Project	Hugging Face links
YuE2 · routed particles	Project / weights · Live demo · Listening gallery
Music 3 · LM LoRA	Project / weights · Live demo · Listening gallery

Try these YuE2 samples. Each Off/On pair uses the same neutral caption, lyrics and seed; only slider strength changes. These are short, token-capped excerpts. The full gallery includes half strength and positive-caption references.

Control	Matched MP3 samples
Female voice	Off · On (+1)
Metal	Off · On (+1)
House	Off · On (+1)

Both projects publish 16 voice and genre controls, with four matched Off/On comparisons per control. YuE2's original particle weights and its ordinary distilled LoRAs are different artifacts: the demo and samples above use the original particles. See the distillation guide for standard LoRA loading and the approximation it introduces.

YuE2: the lead formulation

The learned particles, paired adversarial loss, gradient cap and particle variance/covariance regularizer draw on ParticleGAN, pinned to revision 441fdf42 for this release. The routed transformer adapter is this fork's YuE2 integration.

A frozen YuE2 teacher reads a positive caption. The adapted model reads the neutral caption and learns to reproduce the teacher's prompt-state behavior. Only branches on the AR model's 112 q/k/v/o projections train; the base AR, NAR and VAE weights stay frozen. The published release trains on four caption pairs and reads the final prompt state at the start of music generation.

Terms and notation

Term	Meaning here
Teacher / student	The frozen base model on the positive caption / the adapted model on the neutral caption. A hidden state is the model's internal feature vector.
AR / NAR / VAE	Autoregressive token generation / non-autoregressive decoding / variational autoencoder for audio. These base-model components remain frozen.
LoRA / rank	Low-rank adaptation / the dimension of its small internal bottleneck, here 8.
Particle / router / MLP	A learned vector / the network that chooses a mixture of those vectors / a multilayer perceptron, or small feed-forward neural network.
GAN / RpGAN	Generative adversarial network / its relativistic paired form: a critic learns to distinguish paired real and fake inputs while the adapter learns to defeat it.
MSE / feature matching	Mean squared error / a loss that matches critic features. Neither is part of YuE2's particle matching objective.
VIC	Variance and covariance regularization: keep particles spread out without correlated coordinates.
EMA	Exponential moving average of trained parameters; the smoothed weights used for exports.

A routed-particle adapter

Each projection has its own rank-8 down/up projections, router and nonlinear MLP. All projections in one slider share a learned cloud $P\in\mathbb R^{128\times4}$. In row-vector notation:

$$\begin{aligned} u &= xA^\top,\\ q &= \mathrm{router}(u),\\ z &= \mathrm{softmax}(qP^\top/\sqrt4)P,\\ \Delta(x) &= \mathrm{MLP}([u,z])B^\top,\\ f_s(x) &= f_0(x)+s\frac{\alpha}{r}\Delta(x). \end{aligned}$$

Here $x$ is the layer input, $A$ compresses it to feature $u$ of rank $r$, and $B$ projects the correction back to the layer's output size. The router produces query $q$; softmax turns its particle scores into nonnegative weights that sum to 1, giving the mixture $z$. Square brackets $[u,z]$ concatenate the two vectors; $\top$ means transpose. $f_0$ is the frozen layer, $\Delta$ its learned correction, and $f_s$ the result at slider strength $s$. $\alpha/r$ sets the adapter's overall scale.

The input chooses a soft mixture of particles, which conditions the learned correction. The up projection starts at zero; $r=\alpha=8$. Strength $s=0$ bypasses the adapter exactly, and $s=1$ applies the trained edit. Intermediate strengths interpolate its contribution; negative strengths are untrained. Routing remains active at inference, so these weights cannot be merged into an ordinary $BA$ LoRA. A distilled LoRA is a separately trained approximation.

The paired-error game

Let $h_\theta$ be the adapted neutral-caption state and $h_+$ the frozen positive-caption state. A fixed normalization $T$ puts their error into critic coordinates. Real and fake examples share the same Gaussian noise:

$$\begin{aligned} e &= T(h_\theta)-T(h_+),\\ n &\sim\mathcal N(0,\sigma_t^2 I),\\ x_r &= n,\\ x_f &= n+e. \end{aligned}$$

At exact teacher matching, $e=0$ and the critic sees identical paired inputs. Using $\mathrm{sp}(z)=\log(1+e^z)$, the relativistic paired game is

$$\begin{aligned} \mathcal L_D &= \mathbb E[\mathrm{sp}(D(x_f)-D(x_r))]\\ &\quad +\mathcal R_{\mathrm{cap}},\\ \mathcal L_G &= \mathbb E[\mathrm{sp}(D(x_r)-D(x_f))]\\ &\quad +\mathcal L_{\mathrm{VIC}}(P). \end{aligned}$$

$G$ denotes the trainable adapter and particle cloud; $D$ is the scalar-valued critic (discriminator). They minimize their respective losses $\mathcal L_G$ and $\mathcal L_D$. $\mathbb E$ denotes an average over sampled pairs, and $\mathrm{sp}$ is the smooth positive function softplus. $x_r,x_f$ mean real/fake critic inputs, $n$ is shared Gaussian noise, $\sigma_t$ its standard deviation at update $t$, and $I$ the identity matrix.

The matching signal is purely adversarial. There is no additional output reconstruction MSE, feature matching, lyric hold or ending loss. The two regularizers act on the critic's input gradients and the particle cloud:

$$\begin{aligned} g_x &= \lVert\nabla_xD(x)\rVert_2,\\ \mathcal R_{\mathrm{cap}} &= \tfrac12\sum_{x\in\{x_r,x_f\}}\!\mathbb E[ \max(g_x-1,0)^2]. \end{aligned}$$

The gradient $\nabla_xD(x)$ measures how much the critic score changes with its input; $\lVert\cdot\rVert_2$ is Euclidean length. The cap penalizes only gradient lengths above 1, keeping the critic's sensitivity bounded by a soft penalty.

For the sample covariance $C$ of 64 particles drawn without replacement, with particle dimension $d=4$:

$$\begin{aligned} \mathcal L_{\mathrm{VIC}}(P) &= \frac1d\sum_j\max(0,1-\sqrt{C_{jj}+10^{-4}})\\ &\quad +\frac1d\sum_{j\ne k}C_{jk}^2. \end{aligned}$$

$C_{jj}$ is the sample variance of coordinate $j$; $C_{jk}$ is the covariance between different coordinates $j$ and $k$. The small $10^{-4} stabilizes the square root. The covariance uses the sample denominator \64-1$.

This encourages each particle coordinate to retain variance while discouraging correlation between coordinates. Both regularizers have coefficient 1. In the published release, each update uses separate D/G minibatches of 64 noise-row pairs; the gradient cap runs every fourth update with a factor of 4. Exports use EMA with decay 0.995.

Published release and current experiments

The September 17, 2026 YuE2 release exports the final EMA at 1,200 updates for each of its 16 controls. Its normalization uses the mean and coordinatewise sample standard deviation of frozen positive states, and its noise schedule retains the original 8,000-update horizon:

$$\begin{aligned} T(h) &= \frac{h-\mu_+}{\max(\mathrm{std}(h_+),10^{-4})},\\ \sigma_t &= 0.03^{\min(t/8000,1)}. \end{aligned}$$

Here $\mu_+$ and $\mathrm{std}(h_+)$ are fixed training-set statistics; the maximum is coordinatewise and prevents division by a nearly zero deviation.

The release includes 64 matched Off/On comparisons; checkpoints were exported at the fixed training budget, without an audio-quality selection pass. See the full release math, training audit and stability evidence.

Current training defaults are a later experiment. They use paired-edit normalization, generated continuation histories and a run-budget noise schedule; alternative critics and noise holds are also available. The core paired-error game remains, but a fresh run is not an exact replay of the published release. Implementation: YuE2 adapter and trainer, shared game, reference audit.

Training evidence

The hip-hop run from the September 18 gmix catalog completed 1,600 updates with strong teacher alignment at strength +1. These graphs archive the actual dashboard data in this repository, so viewing them does not require access to that host. This later experiment uses paired-edit normalization, 128 continuation seeds per caption template and a gmix critic: a learned global projection of the error vector into 8 tokens, followed by one attention layer of width 48. It uses 8 noise-row pairs per training minibatch.

Reading the training curves. Generator total is the adversarial loss plus particle VIC; discriminator total includes its gradient cap. Cosine measures alignment between the adapter's hidden-state edit and the positive teacher's edit: 1 means the same direction, 0 means perpendicular, and −1 means opposite. The final training cosine is 0.995. Every update is plotted without smoothing.

Reading the held-out curves. These probes use continuation seeds excluded from training and the EMA weights, sampled every 100 updates at strength +1. Residual RMS is the root mean square of the normalized student–teacher error; p95 is the 95th percentile of per-example RMS errors. SWD means sliced Wasserstein distance: average distribution mismatch along 256 fixed random projections. Teacher SWD compares student and teacher edits; game SWD compares noise with noise-plus-error at a fixed noise standard deviation of 1. Lower is better for all four plotted probe metrics. Gain measures how much of the teacher edit the student produces along its direction; at strength +1, the target is 1.

Held-out metric at +1	Update 100 → 1600
Residual RMS	0.997 → 0.144
Residual p95	1.583 → 0.216
Teacher SWD	0.663 → 0.050
Game SWD, noise std = 1	0.360 → 0.048
Mean gain, target 1	0.183 → 0.989

This is evidence of successful teacher matching for one completed run at +1. Intermediate calibration is still imperfect: strength 0.5 has gain 0.778 at the last probe. The dashboard reports not_plateaued, and its independent classifier evaluation is pending. Audio quality needs listening; the public samples above belong to the separate 1,200-update release.

Archived metrics, exact settings and reproducible plots include all recorded strengths, with no probe points removed.

Models and status

These are separate backends. A checkpoint belongs to its base model, adapter host and training recipe; adapter formats are not interchangeable.

Backend / guide	Adapter host and status
YuE2 particles	AR attention; NAR and VAE frozen. Preferred formulation; published 16-control release.
Music 3 LM	Qwen3 attention for voice and composition. Published 16-control release.
Music 3 flow	Acoustic flow transformer. Earlier mix and production controls.
Music 3 particles	Nonlinear branches on Music 3. Experimental transfer of the particle game.
Krea 2	Text encoder and/or DiT. Opt-in images.
Anima	Conditioner or DiT. Opt-in images.
Supra2-IMG	Cross-attn or DiT on SupraLabs/Supra2-IMG. Opt-in images.
Z-Image Turbo	DiT attention. Opt-in images.
Sana 0.6B	Cross-attention or LoRA. Small image experiments.
LTX-2.5	Text encoder and video connectors. Opt-in video.
MiniMax-H3	Omni-Transformer. Opt-in video and audio.
Tiny LLM	Qwen3-0.6B attention. Particle-bridge test target.
Bonsai GGUF	Frozen readouts. Opt-in particle experiment.
Legacy images	SD 1.x/2.x, SDXL, SD3, Flux and Stable Cascade trainers.

The September 16, 2026 Music 3 release contains 16 unipolar rank-8 LM adapters: female, male, lo-fi, pop, hip-hop, R&B, indie rock, pop punk, metal, country, acoustic folk, house, disco funk, K-pop, reggaeton and afrobeats. It includes native weights, ComfyUI conversions and 64 matched Off/On comparisons. Selected checkpoints range from 1,000 to 3,400 updates; the release card records the per-control choices and limitations.

Other experiments have separate validation records. A successful CPU test, lower training loss or completed render does not establish useful musical control. Reward-model experiments also live here and have their own methods and acceptance checks.

Other training formulations

Music 3's released LM sliders and the inherited image/flow methods use the objectives below. These explain the other backends and the evolution of this fork; YuE2's routed-particle game is defined above.

A strength-controlled adapter

For a linear layer with frozen weight $W_0$, ordinary LoRA learns factors $A\in\mathbb R^{r\times d_{in}}$ and $B\in\mathbb R^{d_{out}\times r}$:

$$W(s)=W_0+s\frac{\alpha}{r}BA.$$

$r$ is the rank, $\alpha$ is the adapter normalization and $s$ is the slider strength. The up projection starts at zero. At $s=0$ the adapter contributes nothing; at $s=1$ it applies the learned unit edit. The complete generator can respond nonlinearly even though this weight update is linear. See lora.py.

Unipolar recipes teach neutral → positive at $+1$. Negative strengths are untrained extrapolation, not a learned opposite. Bipolar recipes explicitly teach both signs. Published female and male controls, for example, are separate unipolar adapters.

At inference the slider operates on the neutral caption. Changing to the positive caption with the adapter off is a teacher reference, a different treatment from applying the slider.

Diffusion and flow targets

Let $f_0(x_t,t,c)$ be the frozen model's prediction under caption $c$, with neutral, positive and negative captions $c_0,c_+,c_-$. A bipolar axis target is

$$\begin{aligned} a_t &= g\big[f_0(x_t,t,c_+)-f_0(x_t,t,c_-)\big],\\ \widehat f_s &= f_0(x_t,t,c_0)+s\,a_t. \end{aligned}$$

The student sees $c_0$ and learns to match $\widehat f_s$. The original image formulation acts on noise predictions. Music 3's acoustic trainer acts on flow velocities and, by default, normalizes its fitting loss:

$$\mathcal L_{\mathrm{NMSE}}= \frac{\mathrm{MSE}\big(f_\theta(x_t,t,c_0;s),\widehat f_s\big)} {\max\big(\mathrm{mean}[(s a_t)^2],10^{-8}\big)}.$$

Its training inputs are anchored to generated clean latents, $x_t=(1-t)\epsilon+t x_0$, using Music 3's noise-to-clean time convention. The alternative pole target learns each caption's displacement from neutral instead of forcing symmetric movement along $c_+-c_-$. Executable definitions: slider_targets.py.

Language-model targets and the common component

Music generation also depends on the autoregressive model that plans the composition. Let $h_0,h_+,h_-$ be its frozen prompt states. Decompose the pair as

$$\begin{aligned} a &= \tfrac12(h_+-h_-),\\ b &= \tfrac12(h_++h_-)-h_0,\\ h_\pm &= h_0+b\pm a. \end{aligned}$$

The older v9 target $h_0\pm a$ discards $b$, the information shared by both pole captions beyond the neutral caption. A perfectly fitted symmetric axis can therefore miss the states occupied by either real caption. Faithful-pole and unipolar targets retain that information. Lyric-token holds, role-specific targets and declared leakage directions address different preservation problems.

See the target geometry, positive/neutral formulation and lyric preservation study. The generic LM trainer still defaults to --lm_target v9 --pole_mode hidden; those defaults do not reproduce the published adversarial release.

The released Music 3 span-GAN objective

The released recipe compares corresponding lyric-token states plus the audio-start state. Frozen neutral and positive sequences provide $H_0,H_+$; the adapted model on the neutral caption provides $H_\theta$. With fixed teacher-RMS calibration $\sigma$:

$$x_+=(H_+-H_0)/\sigma,\qquad x_\theta=(H_\theta-H_0)/\sigma.$$

A transformer critic $D$ learns a relativistic paired comparison. Write $\mathrm{sp}(z)=\log(1+e^z)$:

$$\mathcal L_D=\mathbb E[\mathrm{sp}(D(x_\theta)-D(x_+))] +\mathcal R_{\mathrm{cap}},$$ $$\begin{aligned} \mathcal L_G &= \mathbb E[\mathrm{sp}(D(x_+)-D(x_\theta))]\\ &\quad +\mathcal L_{\mathrm{FM}}+\mathcal L_{\mathrm{end}}. \end{aligned}$$

$\phi$ denotes critic features. Feature matching compares batch means.

$$\begin{aligned} \mu_\theta &= \mathbb E[\phi(x_\theta)],\\ \mu_+ &= \mathbb E[\phi(x_+)],\\ \mathcal L_{\mathrm{FM}} &= \mathrm{MSE}(\mu_\theta,\mu_+). \end{aligned}$$

The one-sided input-gradient cap uses $g_x=\lVert\nabla_xD(x)\rVert_2$ in calibrated critic coordinates:

$$\mathcal R_{\mathrm{cap}}=\frac{\lambda}{2} \sum_{x\in\{x_+,x_\theta\}} \mathbb E\!\left[\max(g_x-\kappa,0)^2\right].$$

The release uses $\lambda=\kappa=1$.

Ending supervision matches the base model's end-versus-continuation margin on the same base-generated token history:

$$\begin{aligned} m &= \ell_{\mathrm{audio\_end}}-\log\sum_{j\in\mathcal S}\exp(\ell_j),\\ \mathcal L_{\mathrm{end}} &= \mathrm{MSE}(m_\theta,m_0), \end{aligned}$$

where $\mathcal S$ is the semantic-token band. All three generator terms have coefficient 1. Explicit lyric hold and direct hidden-state MSE are disabled in this release. The first 600 updates use four rows per batch and fixed base histories. Continuation uses one row per update in balanced shuffled passes, with a fresh base history for ending supervision and a per-parameter update norm cap of 2. Prompt-state teachers remain fixed. See the release formulation and gan_v2/ for implementation details.

Getting started

git clone https://github.com/mikkel/sliders-conceptmod.git
cd sliders-conceptmod

Choose an environment for the backend you intend to use. Do not install the root requirements.txt into a Music 3, YuE2 or modern image/video environment. It is the inherited diffusion-era dependency list, not a universal installer. Base-model weights and their compatible runtimes are separate prerequisites.

On the shared music workstation, train and render on physical GPU 1 while the studio uses GPU 0. CUDA_VISIBLE_DEVICES=1 exposes that card as logical cuda:0; use --device 0 or --device cuda:0 according to the script. The examples below follow this assignment.

YuE2

YuE2 uses its native runtime in a separate environment. The integration targets this pinned upstream revision:

uv venv --python 3.12 .venv-yue2
uv pip install --python .venv-yue2/bin/python \
  'yue2-infer @ git+https://github.com/multimodal-art-projection/YuE.git@ef1936f2ee39fe8de486a0f47a481c95f8d4da87' \
  PyYAML pytest

For ready-to-use controls, start with the original particle weights and the live demo. The native renderer below accepts the particle format directly.

To train a routed-particle slider with the current experimental defaults:

CUDA_VISIBLE_DEVICES=1 .venv-yue2/bin/python \
  conceptmod/textsliders/train_lora_yue2_arm_b.py \
  --recipe particle_bridge \
  --prompts_file conceptmod/textsliders/data/prompts-yue2-metal-arm-b.yaml \
  --save_dir models/metal-yue2-particles --steps 1200

This uses the later paired-edit normalization and continuation-history recipe described above; it does not replay the September 17 release. The trainer expects cached model weights, or a local directory passed as --model_id. Training needs the composition model and tokenizer; rendering also needs the VAE. The renderer below accepts --allow_hub to permit downloads, plus --model_id and --vae_id for local model directories.

To compare an exported adapter, put original section-tagged lyrics in lyrics.txt, then substitute the exported weight path:

CUDA_VISIBLE_DEVICES=1 .venv-yue2/bin/python conceptmod/textsliders/infer_yue2.py \
  --weights /path/to/adapter.safetensors \
  --style 'English, piano-led pop, clear close lead vocal, steady bass and dry drums.' \
  --lyrics_file lyrics.txt --scales=0,0.5,1 --seed 7 \
  --output_dir eval/listen/yue2-example --allow_hub

The renderer preserves matched inputs and records model identity, weight hash, scale and truncation flags. Use a fresh output directory for another run. The full guide covers exact resume, recipe variants, native adapter loading, generation modes and runtime limits.

MiniMax Music 3

Use a working Music 3 environment with the MiniMax Music 3 pipeline and model classes available in its compatible Diffusers installation. The local studio environment is minimax-music3. The acoustic trainer loads from a local model directory; override --model_dir when your weights are elsewhere.

For published adapters, use the native loading example or the ComfyUI guide. Keep each native .safetensors with its JSON sidecar. Start with one adapter at strengths 0 and 1, holding caption, lyrics and seed fixed. The released LM adapters apply to the text encoder/CLIP side in ComfyUI.

To train an acoustic flow slider using the included energy prompts:

conda activate minimax-music3
CUDA_VISIBLE_DEVICES=1 python conceptmod/textsliders/train_lora_music3.py \
  --name energy-example \
  --model_dir /path/to/MiniMax-Music3 \
  --prompts_file conceptmod/textsliders/data/prompts-energy-tf-v7.yaml \
  --save_dir models/energy-example \
  --rank 8 --alpha 8 --steps 500 --seed 7 --device 0

This uses the acoustic trainer's full targets, anchored latents and NMSE defaults. It does not train the published voice/genre LM recipe. For LM training and the release campaign, start with MUSIC3.md, the warm-up campaign and the fresh-continuation campaign. Campaign scripts retain local model/cache paths, manifests and recovery-state requirements; they are research records, not a portable one-command installer.

Use the recipe-comparison pipeline for matched acoustic training/rendering sweeps. Music 3's nonlinear particle experiment has a separate entry point, train_lora_music3_particle.py.

Images and video

Follow the backend's guide in the model table. Each guide specifies its adapter host, teacher space, environment and sampling settings. Several backends expose --dummy for CPU integration checks. Legacy image inference notebooks remain at the repository root; paired-image training lives in trainscripts/imagesliders/.

Evaluation

A useful comparison keeps the neutral caption, lyrics, seed and generation settings fixed, and changes only adapter strength. Include an adapter-off positive-caption reference to show what the base model can do with that prompt. Retain failures, natural endings and truncation flags in the comparison.

Assess concept movement alongside audio quality, lyric preservation, unintended changes and behavior across held-out prompts/seeds. Hidden-state cosine or a GAN loss alone cannot answer those questions. Acoustic ladder gates and LM composition checks measure different behavior; see SCORING.md, LM-SCORING.md and the listening/selection tools.

YuE2's published particle checkpoints use the final 1,200-update EMA, with no audio-quality selection pass. Its listening gallery is the place to judge the control and its side effects; preference for the formulation is not a controlled quality ranking against Music 3 or every other recipe.

The published Music 3 checkpoint policy compares enjoyment and production quality separately, keeps candidates within 0.2 of each best clean mean, and prefers the later qualifying checkpoint. Style similarity and lyric scores are diagnostics, not terms in that selection rule. Four short matched clips per candidate support a shortlist, not a claim of full-song or stacked-adapter reliability. The release card retains the full selection record.

Describe training concepts through sound: instruments, playing, vocal register, breath, mic distance, room and timing. Never put real artist, band, songwriter, producer or album names in prompts, lyrics, titles, listening notes or checkpoint sidecars. Checkpoints trained on named references must be retired and retrained from sound-only prompts.

Development and verification

From the repository root, in a compatible environment with PyTorch, PyYAML, Safetensors and pytest, this CPU suite exercises target geometry, the shared adversarial core and particle mechanics:

CUDA_VISIBLE_DEVICES='' HF_HUB_OFFLINE=1 python -m pytest -q \
  tests/test_2d_slider_geometry.py \
  tests/test_lm_gan.py \
  tests/test_yue2_particle_bridge.py

Native YuE2 tests additionally require its runtime; optional-runtime cases can skip when it is absent. See each backend guide for its tests and dummy training commands. The full historical suite also exercises studio integrations and campaign artifacts: it needs additional dependencies such as SciPy, the parent music workspace's app package for those integrations, and local campaign fixtures. Native GPU training, listening and campaign reproduction require the corresponding models and recovery artifacts.

Repository map

Source	Purpose
Backends and trainers	Model loading, adapters, objectives and inference
GAN engine	Span critics, game updates and recovery states
Reward research	Reward-guided adapters and acceptance checks
Geometry fixtures	Small distribution and objective studies
Campaigns	Source, audits and notes; some need local artifacts
Comparison pipeline	Matched acoustic recipes and render gates
Listening tools	Listening, features and selection experiments
Scripts	Evaluation, dashboards and packaging
Documentation	Backend guides, math and release-card sources
Tests	CPU contracts and backend integration

Local run outputs go in models/, cache/ and eval/listen/. Published weights and recordings are distributed on the Hub.

Lineage and license

The original Concept Sliders paper and implementation introduced LoRA-based concept control for diffusion models. Cite that work when building on it:

@article{gandikota2023sliders,
  title={Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Models},
  author={Rohit Gandikota and Joanna Materzy\'nska and Tingrui Zhou and Antonio Torralba and David Bau},
  journal={arXiv preprint arXiv:2311.12092},
  year={2023}
}

This fork's music, adversarial, particle and newer image/video extensions are documented in the linked source and experiment records. The repository retains the upstream MIT license. Base models, runtimes and vendored code retain their respective licenses; this license does not relicense model weights.