Tutti

anima

flux2

FLUX.2 [dev] is Black Forest Labs' second-generation open-weights image model, released in November 2025. It is a much larger flow-matching transformer than FLUX.1 that uses the Mistral Small 3.1 vision-language model as its text encoder,…FLUX.2 [dev] is Black Forest Labs' second-generation open-weights image model, released in November 2025. It is a much larger flow-matching transformer than FLUX.1 that uses the Mistral Small 3.1 vision-language model as its text encoder, which gives it far better understanding of long, descriptive prompts, reliable multi-line typography and real-world knowledge about brands, materials and lighting. It was designed from the start as a unified model: the same weights generate from text, edit an existing picture from instructions, and combine up to ten reference images for character, product or style consistency, with output up to 4 megapixels. The family also includes the smaller FLUX.2 Klein 4B and 9B distillations that most Flux2 LoRAs in this catalog are trained on. This entry is the text-to-image workflow, accelerated with a turbo LoRA so that it finishes in a handful of steps.Mostra di più

flux

qwen

Qwen-Image is the image generation foundation model of Alibaba's Qwen team, released as open weights in August 2025. It is a 20-billion-parameter multimodal diffusion transformer that reads prompts through the Qwen2.5-VL vision-language…Qwen-Image is the image generation foundation model of Alibaba's Qwen team, released as open weights in August 2025. It is a 20-billion-parameter multimodal diffusion transformer that reads prompts through the Qwen2.5-VL vision-language model, and its headline strength is text: it renders long paragraphs, multi-line signs, posters and mixed Chinese and English typography with an accuracy no earlier open model reached. The same team's technical report also shows strong results for layout-heavy prompts, infographics and precise image editing, and the companion Qwen-Image-Edit model is the basis of many editing workflows. It generates natively around 1.3 to 1.6 megapixels in the usual aspect ratios and takes natural-language prompts rather than tag lists; the team recommends appending quality phrases such as 'Ultra HD, 4K, cinematic composition'. This entry is the version 2512 text-to-image workflow (the December 2025 refresh of the weights); the original release is kept as qwen-v1.Mostra di più

sdxl

Stable Diffusion XL version 1.0 is here, and the Refiner 1.0 is also installed. This is the official full release, but our software needs adjustment to fully support it. Only the /render command works and SDXL cannot be mixed with old…Stable Diffusion XL version 1.0 is here, and the Refiner 1.0 is also installed. This is the official full release, but our software needs adjustment to fully support it. Only the /render command works and SDXL cannot be mixed with old styles, recipes, loras, inversions, and other commands yet. Keep an eye on the XL news topic in Telegram VIP for more informationMostra di più

DeepSeek V4.1 Flash

DeepSeek | 1,048,576 token context | $0.04/M input, $0.68/M output. DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek, and the first built on the company's Causal Encoder-Decoder (CED) architecture. It activates 8B…DeepSeek | 1,048,576 token context | $0.04/M input, $0.68/M output. DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek, and the first built on the company's Causal Encoder-Decoder (CED) architecture. It activates 8B parameters on input and 16B on output from a 552B-parameter backbone, an asymmetric split that keeps per-token compute low relative to the model's total size. Image understanding is native to the architecture, with visual and text embeddings trained jointly from the start of pre-training rather than added afterward as in the earlier experimental V4 Flash Vision ExpOpens in new tab. It is suited for coding, terminal, and computer-use agents, along with long-horizon tasks that must run to completion across many steps and long-context analysis. Compressed KV caching cuts cache memory to roughly a quarter of the previous Flash generation, significantly reducing costs on agentic workloads. DeepSeek positions it as the cost-efficient tier of the V4.1 family and reports that it exceeds V4 ProOpens in new tab on performance, speed, and task completion time.Mostra di più

sdv15

Stable Diffusion 1.5 by Stability / RunwayML, the OG model. Someone said it was weird that we didn't have this in the concepts collection, and they were correct, so here you go! Of course, this model is over a year old, but it's still…Stable Diffusion 1.5 by Stability / RunwayML, the OG model. Someone said it was weird that we didn't have this in the concepts collection, and they were correct, so here you go! Of course, this model is over a year old, but it's still fantastic in its own right.Mostra di più

zimage turbo

ReV Animated

Counterfeit V3.0

CounterfeitV3.0 by rqdwdw. The 3.0 version of Counterfeit has much brighter colors and more focus on the subject, whereas the previous model was more muted and pastel-like. The backgrounds look totally different in particular, so we're…CounterfeitV3.0 by rqdwdw. The 3.0 version of Counterfeit has much brighter colors and more focus on the subject, whereas the previous model was more muted and pastel-like. The backgrounds look totally different in particular, so we're going to keep both. It should be noted that there are 10 upvotes and 3 comments about green spots from users on Civitai.Mostra di più

Wan 2.1

Wan 2.1 is the February 2025 generation of Alibaba's open Wan video models: 1.3B and 14B weights for text-to-video and image-to-video. It has been superseded by Wan 2.2 (July 2025), which gives clearly better motion, detail and prompt…Wan 2.1 is the February 2025 generation of Alibaba's open Wan video models: 1.3B and 14B weights for text-to-video and image-to-video. It has been superseded by Wan 2.2 (July 2025), which gives clearly better motion, detail and prompt following with the same kind of workflows, so there is no reason to start with Wan 2.1 in 2026 or later. This page stays as a reference for the older LoRAs and workflows that were made for it.Mostra di più

GLM 5.3 Flash

Z.ai | 1,048,576 token context | $0.02/M input, $0.72/M output. GLM-5.3-Flash is a native multimodal model from Z.ai. It is suited for efficient coding and long-horizon agent tasks. Its hybrid sparse and linear attention architecture…Z.ai | 1,048,576 token context | $0.02/M input, $0.72/M output. GLM-5.3-Flash is a native multimodal model from Z.ai. It is suited for efficient coding and long-horizon agent tasks. Its hybrid sparse and linear attention architecture maintains accurate long-context behavior while reducing compute overhead. Yes, the stealth model Ox Alpha was revealed to be ZAI's new model, GLM-5.3 Flash.Mostra di più

Hy4 preview

Tencent | 1,048,576 token context | $0.83/M input, $2.50/M output. Tencent: Hy4 preview is a mixture-of-experts model from Tencent, with 49B active parameters out of 770B total. It is designed for coding agents, complex tool-use workflows,…Tencent | 1,048,576 token context | $0.83/M input, $2.50/M output. Tencent: Hy4 preview is a mixture-of-experts model from Tencent, with 49B active parameters out of 770B total. It is designed for coding agents, complex tool-use workflows, and productivity tasks that require planning, context continuity, and sustained multi-step execution.Mostra di più

MiMo V2.6 Flash

Xiaomi | contesto da 1.050.000 token | $0.14/M input, $0.28/M output. MiMo-V2.6-Flash è un modello foundation open-source sviluppato da Xiaomi. Costruito su un'architettura Mixture-of-Experts con 309B parametri totali e 15B attivati per…Xiaomi | contesto da 1.050.000 token | $0.14/M input, $0.28/M output. MiMo-V2.6-Flash è un modello foundation open-source sviluppato da Xiaomi. Costruito su un'architettura Mixture-of-Experts con 309B parametri totali e 15B attivati per token, impiega un meccanismo di attenzione ibrido per una maggiore efficienza computazionale. Il modello presenta una finestra di contesto da 1M token e capacità multimodali native. Ottimizzato per flussi di lavoro agentici, offre prestazioni solide in coding, visione, scenari generali e di ricerca, eccellendo in compiti complessi a lungo termine con robusta generalizzazione su una vasta gamma di harness agentici.Mostra di più

DeepSeek V4 Flash 0731

DeepSeek | 1,048,576 token context | $0.01/M input, $1.28/M output. DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for…DeepSeek | 1,048,576 token context | $0.01/M input, $1.28/M output. DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows. This is the GA release of DeepSeek V4 Flash.Mostra di più

lightning

Detail Tweaker XL

Nemotron 3 Ultra

NVIDIA | contesto da 262.144 token | $0.60/M input, $2.40/M output. NVIDIA Nemotron 3 Ultra è un modello open frontier-reasoning e orchestrazione di NVIDIA, con 55B parametri attivi su 550B totali (MoE). Costruito su un'architettura ibrida…NVIDIA | contesto da 262.144 token | $0.60/M input, $2.40/M output. NVIDIA Nemotron 3 Ultra è un modello open frontier-reasoning e orchestrazione di NVIDIA, con 55B parametri attivi su 550B totali (MoE). Costruito su un'architettura ibrida Transformer-Mamba mixture-of-experts, supporta input e output testuali con una finestra di contesto fino a 1M token. È adatto per flussi di lavoro agentici a lunga durata, inclusa orchestrazione agentica, agenti di coding, ricerca approfondita e compiti aziendali complessi. È particolarmente forte nel ragionamento e pianificazione multi-step, con inferenza ad alto throughput progettata per pipeline agentiche ad alto volume. Fa parte della famiglia NVIDIA Nemotron di modelli aperti per AI agentica.Mostra di più

Step 5 Preview

StepFun | 1,048,576 token context | $1.00/M input, $2.70/M output. Step 5 Preview is StepFun's flagship model for agentic work, built on a sparse Mixture-of-Experts architecture (27B active / 600B total parameters). It performs strongly in…StepFun | 1,048,576 token context | $1.00/M input, $2.70/M output. Step 5 Preview is StepFun's flagship model for agentic work, built on a sparse Mixture-of-Experts architecture (27B active / 600B total parameters). It performs strongly in software engineering and professional knowledge work, with particular strength in finance. It is designed for extended tasks that span large codebases and documents, using tools and refining results over multiple steps.Mostra di più

Add More Details Detail Enhancer / Tweaker (细节调整) LoRA

Add More Details - Detail Enhancer / Tweaker (细节调整) LoRA by Lykon writes: his between 0.5 and 1 weight, depending on your preference. You can go lower than 0.5 for a more subtle effect, of course. I also found out that this gives some…Add More Details - Detail Enhancer / Tweaker (细节调整) LoRA by Lykon writes: his between 0.5 and 1 weight, depending on your preference. You can go lower than 0.5 for a more subtle effect, of course. I also found out that this gives some interesting results at negative weight, sometimesMostra di più

waifu

waifu-diffusion by hakurei. A long running base model trained on crisp-looking anime. Compared to your other anime base model options, this one is fairly well-rounded, which unfortunately has been detrimental in making it a clear winner in…waifu-diffusion by hakurei. A long running base model trained on crisp-looking anime. Compared to your other anime base model options, this one is fairly well-rounded, which unfortunately has been detrimental in making it a clear winner in any category.Mostra di più

hyper

GLM 5.3

Z.ai | contesto da 1.048.576 token | $0.08/M input, $4.80/M output. GLM-5.3 è un modello di ragionamento su larga scala di Z.ai, costruito per ingegneria del software complessa e compiti agentici a lungo termine. Supporta input e output…Z.ai | contesto da 1.048.576 token | $0.08/M input, $4.80/M output. GLM-5.3 è un modello di ragionamento su larga scala di Z.ai, costruito per ingegneria del software complessa e compiti agentici a lungo termine. Supporta input e output testuali con una finestra di contesto da 1M token, e migliora GLM-5.2 nel coding e nell'equilibrio tra prestazioni ed efficienza dei token. Il ragionamento è sempre attivo e non può essere disabilitato. Supporta sforzi di ragionamento low, high e max; max è il valore predefinito.Mostra di più

krea

Detail Tweaker LoRA (细节调整LoRA)

LoRA designed for removing or adding detail while keeping the style of the render the same. Current strength is good from 0 - 2 depending on your use case. In the future, it can be used as negative weight to reduce details. Works with…LoRA designed for removing or adding detail while keeping the style of the render the same. Current strength is good from 0 - 2 depending on your use case. In the future, it can be used as negative weight to reduce details. Works with anime, semi-real and realistic models and styles.Mostra di più

DeepSeek V4 Flash 0423

DeepSeek | contesto da 1.048.576 token | $0.09/M input, $0.18/M output. DeepSeek V4 Flash è un modello Mixture-of-Experts ottimizzato per l'efficienza di DeepSeek con 284B parametri totali e 13B parametri attivi, supportando una finestra…DeepSeek | contesto da 1.048.576 token | $0.09/M input, $0.18/M output. DeepSeek V4 Flash è un modello Mixture-of-Experts ottimizzato per l'efficienza di DeepSeek con 284B parametri totali e 13B parametri attivi, supportando una finestra di contesto da 1M token. È progettato per inferenza veloce e carichi di lavoro ad alto throughput, mantenendo forti prestazioni in ragionamento e coding. Il modello include attenzione ibrida per un'elaborazione efficiente di contesti lunghi. Supporta sforzi di ragionamento high e xhigh; xhigh corrisponde al ragionamento massimo. È adatto per applicazioni come assistenti di coding, sistemi di chat e flussi di lavoro agentici dove reattività ed efficienza dei costi sono importanti.Mostra di più

Hy3

Tencent | 262,144 token context | $0.13/M input, $0.53/M output. Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production…Tencent | 262,144 token context | $0.13/M input, $0.53/M output. Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort: a direct no-think mode by default, plus low and high chain-of-thought modes for complex math, coding, and multi-step problems. With a 256K context window, Hy3 targets long-horizon tasks, including improved coreference resolution, multi-turn constraint tracking, and stable tool-calling that generalizes across agent scaffoldings. Tencent positions it as a reliable, cost-effective option across coding, document processing, financial analysis, game development, and frontend design, with a strong emphasis on grounded, anti-hallucination behavior that answers when grounded and flags when evidence is missing rather than fabricating.Mostra di più

ebara_pony

Plant Milk Model Suite Walnut

Kimi K3

MoonshotAI | 1,048,576 token context | $0.30/M input, $11.94/M output. Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic…MoonshotAI | 1,048,576 token context | $0.30/M input, $11.94/M output. Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at navigating large repositories, using tools, debugging, and iterating against images, logs, tests, and runtime feedback. Its architecture uses KDA and Attention Residuals for computational efficiency.Mostra di più

MiMo V2.5

Xiaomi | contesto da 1.050.000 token | $0.14/M input, $0.28/M output. MiMo-V2.5 è un modello omnimodale nativo di Xiaomi. Offre prestazioni agentiche di livello Pro a circa metà del costo di inferenza, superando MiMo-V2-Omni nella…Xiaomi | contesto da 1.050.000 token | $0.14/M input, $0.28/M output. MiMo-V2.5 è un modello omnimodale nativo di Xiaomi. Offre prestazioni agentiche di livello Pro a circa metà del costo di inferenza, superando MiMo-V2-Omni nella percezione multimodale su compiti di comprensione di immagini e video. La sua finestra di contesto da 1M supporta documenti completi, conversazioni estese e contesti di compiti complessi in un unico passaggio, rendendolo ideale per integrazione con framework agentici dove contano ragionamento forte, percezione ricca ed efficienza dei costi.Mostra di più

extremely detailed (no trigger) sliders.ntcai.xyz

GLM 5.2

Z.ai | 1,048,576 token context | $0.04/M input, $9.00/M output. GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, and is suited for long-horizon agent workflows,…Z.ai | 1,048,576 token context | $0.04/M input, $9.00/M output. GLM 5.2 is a large-scale reasoning model from Z.ai. It supports text input and output with a 1M-token context window, and is suited for long-horizon agent workflows, project-level software engineering, and complex multi-step automation. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is particularly strong at coding and tool use across long-running tasks, able to maintain engineering context and follow standards consistently through a full development workflow, from requirements to multi-platform deployment, in a single task.Mostra di più

MiniMax M3

MiniMax | contesto da 1.048.576 token | $0.30/M input, $1.20/M output. MiniMax-M3 è un modello foundation multimodale di MiniMax. Supporta input testuali, immagini e video con output testuale, una finestra di contesto da 1M token, ed è…MiniMax | contesto da 1.048.576 token | $0.30/M input, $1.20/M output. MiniMax-M3 è un modello foundation multimodale di MiniMax. Supporta input testuali, immagini e video con output testuale, una finestra di contesto da 1M token, ed è adatto per lavoro agentico a lungo termine, coding e uso di strumenti. È costruito su MiniMax Sparse Attention (MSA), che sostituisce l'attenzione completa con selezione KV-block per ridurre il calcolo per token su contesti lunghi — circa 1/20 del costo della generazione precedente a 1M token, con prefill e decode sostanzialmente più veloci mantenendo qualità su quasi tutti i compiti. Addestrato come modello multimodale nativo su dati intercalati e ottimizzato per collaborazione multi-turno simile a produzione tramite un framework interattivo con simulatore utente, il modello è orientato a compiti sostenuti e multi-step piuttosto che a esecuzione single-turn.Mostra di più

[KREA 2] Detail Slider

MiMo V2.6 Pro

Xiaomi | contesto da 1.050.000 token | $0.43/M input, $0.87/M output. MiMo-V2.6-Pro è il modello fondazionale di punta sviluppato da Xiaomi. Costruito su una scala di oltre 1T parametri, è progettato per spingere al massimo le capacità per…Xiaomi | contesto da 1.050.000 token | $0.43/M input, $0.87/M output. MiMo-V2.6-Pro è il modello fondazionale di punta sviluppato da Xiaomi. Costruito su una scala di oltre 1T parametri, è progettato per spingere al massimo le capacità per i carichi di lavoro più esigenti. Il modello presenta una finestra di contesto da 1M token e capacità multimodali native. Ottimizzato per flussi di lavoro agentici, offre prestazioni di alto livello in scenari di coding, visivi, generali e di ricerca, eccellendo in compiti complessi e a lungo termine con una robusta generalizzazione su una vasta gamma di agent harness.Mostra di più

ICBINP XL

DeepSeek V4 Pro 0813

Laguna S 2.1

Poolside | contesto da 1.048.576 token | $0.09/M input, $0.18/M output. Laguna S 2.1 è l'ultimo modello per agenti di coding di PoolsideOpens in new tab. Laguna S 2.1 è un modello da 118B parametri totali con 8B parametri attivi, che…Poolside | contesto da 1.048.576 token | $0.09/M input, $0.18/M output. Laguna S 2.1 è l'ultimo modello per agenti di coding di PoolsideOpens in new tab. Laguna S 2.1 è un modello da 118B parametri totali con 8B parametri attivi, che ottiene il 70.2% su Terminal-Bench 2.1 e il 40.4% su DeepSWE, rendendolo uno dei modelli di coding più forti nella sua categoria. Open-weight sotto licenza OpenMDW-1.1. Laguna S 2.1 è progettato per casi d'uso di ingegneria del software e coding agentico, e sei responsabile di confermare che sia appropriato per la tua applicazione prevista. Laguna S 2.1 è soggetto alla licenza OpenMDW-1.1Opens in new tab, e dovrebbe essere usato in conformità con la Acceptable Use PolicyOpens in new tab di Poolside. Si sconsiglia di aggirare le salvaguardie di sicurezza di Laguna S 2.1 senza implementare mitigazioni sostanzialmente equivalenti appropriate per il tuo caso d'uso. Si prega di segnalare vulnerabilità di sicurezza o preoccupazioni sulla sicurezza a [email protected]. Se usi Laguna S 2.1 gratuitamente, potremmo usare i tuoi input e output per addestrare e migliorare i nostri modelli.Mostra di più

lyh_anime_Flux

Background Detail Enhancer

krea2

Nemotron 3.5 Lightning

NVIDIA | contesto da 262.144 token | $0.07/M input, $0.20/M output. NVIDIA Nemotron 3.5 Lightning è un modello open Mixture-of-Experts di NVIDIA, con 3B parametri attivi su 30B totali. È adatto per carichi di lavoro agentici ad alto…NVIDIA | contesto da 262.144 token | $0.07/M input, $0.20/M output. NVIDIA Nemotron 3.5 Lightning è un modello open Mixture-of-Experts di NVIDIA, con 3B parametri attivi su 30B totali. È adatto per carichi di lavoro agentici ad alto throughput e compiti specializzati che beneficiano di personalizzazioni specifiche per dominio.Mostra di più

playground2

DeepSeek V4 Pro 0423

DeepSeek | 1,048,576 token context | $0.96/M input, $1.91/M output. DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It…DeepSeek | 1,048,576 token context | $0.96/M input, $1.91/M output. DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It is designed for advanced reasoning, coding, and long-horizon agent workflows, with strong performance across knowledge, math, and software engineering benchmarks. Built on the same architecture as DeepSeek V4 Flash, it introduces a hybrid attention system for efficient long-context processing. Reasoning efforts high and xhigh are supported; xhigh maps to max reasoning. It is well suited for complex workloads such as full-codebase analysis, multi-step automation, and large-scale information synthesis, where both capability and efficiency are criticalMostra di più

Chroma

gpt oss 120b

OpenAI | contesto da 131.072 token | $0.15/M input, $0.60/M output. gpt-oss-120b è un modello linguistico Mixture-of-Experts (MoE) open-weight da 117B parametri di OpenAI progettato per casi d’uso di produzione ad alto ragionamento,…OpenAI | contesto da 131.072 token | $0.15/M input, $0.60/M output. gpt-oss-120b è un modello linguistico Mixture-of-Experts (MoE) open-weight da 117B parametri di OpenAI progettato per casi d’uso di produzione ad alto ragionamento, agentici e generali. Attiva 5.1B parametri per ogni passaggio forward ed è ottimizzato per funzionare su una singola GPU H100 con quantizzazione MXFP4 nativa. Il modello supporta profondità di ragionamento configurabile, accesso completo alla catena di pensiero e uso nativo di strumenti, inclusi chiamate di funzione, navigazione e generazione di output strutturati.Mostra di più

level4

Level4 by AreThoseLevel4Plates is the creator of and this one, made popular for its stunning photos of athletes. Level4 is great for beginners, because it responds well to simple prompts, but to get a realistic photo out of it may take…Level4 by AreThoseLevel4Plates is the creator of and this one, made popular for its stunning photos of athletes. Level4 is great for beginners, because it responds well to simple prompts, but to get a realistic photo out of it may take more work, and there are better options. As a semi-realistic general model it's efforlessly great for portraits, cars, landscapes. The creator demos it as the go-to model for creating female athletes. Try the #nfix recipe with it or #cinematicMostra di più

Realism Enhancer Krea2

ostrisflex