ModelSFC-Locoformer: adaptive spectral compression at 11.8 dB on DnR
A technical reading of learned spectral compression, local/long-range sequence modelling and the same-protocol 11.8 dB DnR result.
Read article
ModelBoogu-Image 0.1: inference paths and failure modes across Base, Turbo and Edit
Latent flow matching, few-step distillation and edit preservation across three released checkpoints, with runtime and evidence boundaries.
Read article
ModelSenseNova-U1 Infographic V2: the scope of its 50.3 infographic score
Typography, layout and semantics are evaluated separately, with MoT routing and the BizGenEval hard result kept in scope.
Read article
ModelLongCat-AudioDiT 3.5B: conditioning an audio diffusion transformer on a speaker
Conditional diffusion, APG/CFG, 32 function evaluations and the limits of Seed speaker-similarity measurements.
Read article
ModelAudioSep-Hive: turning language into a query for open-world sound separation
CLAP conditioning, time-frequency masking and mixture consistency in one query-driven model trained across 283 sound concepts.
Read article
ModelBandIt and TIGER-DnR: where two three-stem separators spend their compute
A same-protocol comparison of band-split modelling and TIGER's efficient sequence design on DnR cinematic separation.
Read article
ModelMossFormer2 and TIGER-speech: separating speakers is not speech restoration
A task-level analysis of permutation-invariant separation, enhancement and super-resolution across the ClearVoice and TIGER profiles.
Read article
ModelSpecialist audio-to-MIDI: the boundaries of GAPS, Filo, VioPTT and SwiftF0
Corrected output boundaries for GAPS, FiloSax, FiloBass, VioPTT and SwiftF0, plus public-weight candidates and the E1–E5 acceptance gates required before publication.
Read article
ModelMuScriptor-large: evidence and limits of a 1.366B full-mix-to-MIDI model
An evidence-led account of the flagship checkpoint, event modelling, private and cross-domain evaluation, Type-1 MIDI output, and gated non-commercial license.
Read article
ModelQwen-Image, Z-Image and FLUX.2 Klein: why three open image lines diverged
A technical history of the 20B MMDiT, 6B single-stream DiT and 9B rectified-flow routes across typography, distillation and multi-reference editing.
Read article
ModelBiRefNet, Moebius, VOSR and DDColor: four objectives for matting, inpainting, restoration and color
Four different objectives—matting, masked inpainting, generative SR and colour regression—kept as separate product contracts.
Read article
Modeldots.tts: continuous acoustic latents, flow matching and 48 kHz autoregressive speech
A technical walkthrough of AudioVAE, semantic re-encoding, the Qwen backbone and AR flow matching in SOAR and MeanFlow profiles.
Read article
ModelSoulX-Singer and YingMusic-Singer-Plus: score control versus melody reference
Zero-shot timbre prompts, MIDI metadata alignment, melody encoders and conditional flow matching with non-interchangeable inputs.
Read article
ModelHQ-SVC and YingMusic-SVC: disentangling content, F0, energy and timbre
Content encoding, pitch conditioning, speaker statistics and rectified flow in zero-shot singing voice conversion.
Read article
ModelBeat This!: recovering beats and downbeats without a DBN
Multi-resolution spectra, RoFormer sequence modelling and shift-tolerant objectives, with beat timing kept separate from note transcription.
Read article
ModelDemucs v3 for saxophone: 14.22 dB and narrow-domain catalogue A/B testing
Time-domain encoding, sequence bottlenecks and waveform reconstruction, with the reported 14.22 dB result kept inside its narrow jazz test domain.
Read article
ModelQwen3-TTS, Fish S2 Pro, MOSS-TTS and VoxCPM2: four speech-conditioning contracts
Voice design, discrete audio tokens, long-context scaling and zero-shot references compared as distinct inference contracts rather than one score.
Read article
ModelVevo2: separating content, timbre and prosody in zero-shot singing
Content-style and prosody tokenizers, flow-matching acoustics and vocoder reconstruction in a reference-driven singing path.
Read article
ModelFour-stem audio separation: why Huge-SCNet fits the MSS target
TelkNet's current four-stem tool uses Huge-SCNet-4stems V1.2 to split a full mix into vocals, drums, bass, and other. The article explains the choice through MSS architecture history, Mel-Band RoFormer strengths and limits, SCNet subband modeling, and public four-stem scores.
Read article
ModelKrea 2: Krea AI releases Raw and Turbo image-generation weights
Krea AI released Krea 2 Raw and Krea 2 Turbo as open weights. The technical report describes Krea 2 as a 12B text-to-image diffusion model trained from scratch and details data filtering, captioning, post-training, reinforcement learning, and safety-license boundaries.
Read article
ModelChordEdit: one-step low-energy transport for source-to-target image editing
ChordEdit is a CVPR 2026 Oral / Best Student Paper Honorable Mention paper. It treats one-step text-guided image editing as a dynamic optimal-transport problem between source and target prompts, then uses a Chord Control Field to reduce high-energy drift that can warp objects and damage backgrounds. The official demo exposes source_prompt, target_prompt, seed, sample count, time-window, and step-scale controls.
Read article
ModelZyphra ZONOS2: open TTS MoE, voice cloning, and tool-workflow boundaries
Zyphra released ZONOS2 under Apache-2.0. The model uses an 8B-total, roughly 900M-active MoE TTS architecture trained on more than 6M hours of speech. Its official materials document benchmark framing, language tiers, and voice-cloning limits.
Read article
ModelGoogle Magenta RealTime 2: when a music model becomes a playable instrument
A source-scoped technical overview of MRT2's 40ms frame-level autoregression, low-latency MIDI/text/audio control, 2.4B/230M model split, SpectroStream codec path, MusicCoCa style embeddings, and the practical boundary between live steering and one-shot song generation.
Read article