Text-to-Image¶
Image & 2D Art
Open Source¶
-
2025
An 8.9B FLUX.1-schnell-derived, fully Apache-2.0 text-to-image model trained on a broad artistic and photographic dataset, offered as an unrestricted open base for fine-tuning.
apache-2.0diffusionfluxopen-weights -
2025
An 8B single-stream diffusion-transformer text-to-image model from Baidu, released with open weights.
diffusion-transformeropen-weights -
F-Lite Discovered by AI
2025
A 10-billion parameter open-source text-to-image model developed by Freepik and Fal.ai. It is trained exclusively on licensed, copyright-safe, and SFW content, offering high-quality image generation with ethical data sourcing.
copyright-safediffusiontext-to-image -
2025
A 32B-parameter flow-matching model that unifies text-to-image generation, instruction editing, and multi-reference conditioning in one open-weights checkpoint, supporting character and style reference without fine-tuning.
diffusionfluxmulti-referenceopen-weights -
2025
A 16B hybrid autoregressive-plus-diffusion text-to-image model from Zhipu AI, released with open weights.
autoregressivediffusionhybridopen-weights -
2025
A 17B sparse diffusion-transformer text-to-image model released with open weights, targeting state-of-the-art prompt adherence and image quality.
diffusion-transformermixture-of-expertsopen-weights -
2025
An 80B mixture-of-experts text-to-image model from Tencent Hunyuan, one of the largest open-weight image generators, aimed at high-fidelity native-resolution synthesis.
mixture-of-expertsopen-weights -
Janus-Pro Discovered by AI
2025
Janus-Pro is an advanced unified multimodal understanding and generation model by DeepSeek. It features an optimized training strategy, expanded training data, and scaled model sizes, achieving state-of-the-art text-to-image instruction-following and generation stability.
diffusionmultimodaltext-to-imageGitHub · arXiv · Hugging Face
-
2025
A compact 2.6B diffusion-transformer text-to-image model using a Gemma-2-2B text encoder, balancing quality and efficiency for accessible high-resolution generation.
diffusion-transformerefficientopen-weights -
Lumina-Image 2.0: A Unified and Efficient Image Generative Framework Discovered by AI
2025
Lumina-Image 2.0 is a 2.6-billion parameter flow-based diffusion transformer designed for high-quality text-to-image generation. It integrates a unified Next-DiT architecture that treats text and image tokens as a joint sequence, enabling natural cross-modal interactions and seamless task expansion.
diffusionflow-matchingtransformerGitHub · arXiv · Hugging Face
-
OmniGen2: Exploration to Advanced Multimodal Generation Discovered by AI
2025
OmniGen2 is a unified multimodal generation model by BAAI that supports text-to-image synthesis, instruction-based image editing, and subject-driven generation. It features decoupled decoding pathways for text and image modalities to preserve language capabilities while producing consistent visual outputs.
diffusionmultimodalunifiedGitHub · arXiv · Hugging Face
-
2025
Alibaba's open-weight text-to-image family with notably strong in-image text rendering, including Chinese, and support for both generation and editing.
diffusionopen-weightstext-rendering -
SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformer Discovered by AI
2025
An efficiency-oriented linear Diffusion Transformer framework that scales text-to-image generation up to 4K resolution. It introduces depth-growth training, 8-bit optimization, and model depth pruning to enable fast, high-resolution synthesis on consumer-grade GPUs.
diffusionsanatext-to-imagetransformer -
SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillation Discovered by AI
2025
SANA-Sprint is an ultra-fast text-to-image model that achieves high-quality generation in just 1 to 4 steps using continuous-time consistency distillation. Built on the Sana foundation, it dramatically reduces latency, enabling real-time image generation at 1024×1024 resolution.
diffusiondistillationreal-time -
Show-o Turbo Discovered by AI
2025
An accelerated unified multimodal model that leverages consistency distillation and trajectory segmentation to significantly speed up text-to-image generation and multimodal understanding.
diffusionmultimodaltext-to-imagetransformer -
Show-o2: Improved Native Unified Multimodal Models Discovered by AI
2025
Show-o2 is an improved native unified multimodal model that integrates autoregressive modeling and flow matching on a 3D causal variational autoencoder space. It enables scalable, high-quality text-to-image and video generation alongside robust multimodal understanding.
autoregressiveflow-matchingmultimodaltext-to-image -
2025
A 6B text-to-image model family from Alibaba's Tongyi-MAI, designed for efficient single-GPU high-quality generation with strong bilingual text rendering.
diffusionefficientopen-weightstext-rendering -
Chameleon: Mixed-Modal Early-Fusion Foundation Models Discovered by AI
2024
A family of early-fusion token-based mixed-modal models capable of understanding and generating images and text in any arbitrary sequence. It unifies text and image modalities into a single transformer architecture, enabling seamless text-to-image generation and multimodal reasoning.
chameleonmultimodaltext-to-imagetransformer -
Emu3 Discovered by AI
2024
Emu3 is a state-of-the-art native multimodal model suite trained solely with next-token prediction. It tokenizes images, text, and videos into a discrete space, demonstrating powerful text-to-image generation and multimodal understanding capabilities.
autoregressivemultimodaltext-to-image -
2024
A 12B-parameter rectified-flow transformer text-to-image family from Black Forest Labs; the guidance-distilled [dev] and 1-4 step [schnell] variants are a leading open-weights base for concept art and LoRA fine-tuning.
diffusionfluxopen-weightsrectified-flow -
Hunyuan-DiT Discovered by AI
2024
A powerful multi-resolution diffusion transformer with fine-grained bilingual (English and Chinese) understanding and multi-turn dialogue capabilities.
bilingualdiffusiondittext-to-imageGitHub · arXiv · Hugging Face
-
Janus Discovered by AI
2024
An autoregressive framework that unifies multimodal understanding and generation by decoupling visual encoding into separate pathways, achieving high-quality text-to-image generation.
autoregressivedeepseekmultimodaltext-to-image -
Kandinsky 3 Discovered by AI
2024
A text-to-image diffusion model based on latent diffusion, designed for high-quality, photorealistic image generation and easy adaptation to downstream tasks like editing, inpainting, and video generation.
diffusionlatent-diffusiontext-to-image -
Kolors Discovered by AI
2024
Kolors is a large-scale text-to-image generation model based on latent diffusion, developed by Kuaishou. It excels in visual quality, complex semantic accuracy, and text rendering for both Chinese and English prompts.
bilingualdiffusiontext-to-image -
LlamaGen: Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation Discovered by AI
2024
A family of image generation models that applies the vanilla autoregressive next-token prediction paradigm of large language models to the visual generation domain.
autoregressivellamatext-to-image -
Lumina-Next: Making Lumina-T2X Stronger and Faster with Next-DiT Discovered by AI
2024
An upgraded version of the Lumina-T2X framework that utilizes Next-DiT to improve text-to-image generation speed and quality. It addresses training instability and extrapolation artifacts while supporting high-resolution generation with Gemma-2B as the text encoder.
diffusionluminatext-to-imagetransformer -
OmniGen Discovered by AI
2024
A unified image generation model based on a simplified diffusion transformer architecture that natively supports text-to-image, image editing, and subject-driven generation without requiring auxiliary modules like ControlNet or IP-Adapter.
diffusionmultimodaltext-to-imagetransformer -
PixArt-Sigma Discovered by AI
2024
An advanced Diffusion Transformer (DiT) model capable of directly generating high-quality images at 4K resolution using a weak-to-strong training methodology.
4kdiffusiondittext-to-imageGitHub · arXiv · Hugging Face
-
2024
An aesthetics-focused text-to-image model on the SDXL architecture, tuned for color, contrast, and human-centric detail; popular for polished concept-art looks.
aestheticopen-weightssdxl -
2024
An efficient high-resolution (up to 4K) text-to-image model from NVIDIA using a linear diffusion transformer and deep-compression autoencoder, fast enough to run on laptop GPUs.
efficientlinear-diffusion-transformernvidiaopen-weights -
Show-o: One Single Transformer to Unify Multimodal Understanding and Generation Discovered by AI
2024
A unified transformer model that integrates autoregressive and discrete diffusion modeling to support both multimodal understanding and high-quality text-to-image generation.
diffusionmultimodaltext-to-imagetransformer -
Stable Diffusion 3.5 Discovered by AI
2024
An advanced family of open-weights text-to-image diffusion models featuring improved image quality, typography, complex prompt understanding, and resource efficiency.
diffusionstability-aitext-to-imagetransformer -
2024
An 8B MMDiT text-to-image model with three text encoders and rectified-flow training, Stability AI's open-weight flagship for high-resolution, prompt-faithful generation.
mmditopen-weightsrectified-flowGitHub · arXiv · Hugging Face
-
2023
A cascaded pixel-space diffusion model (64 to 256 to 1024px) with strong prompt and in-image text fidelity, useful when legible text such as signage or UI mockups matters.
cascadeddiffusionpixel-spacetext-rendering -
2023
An open text-to-image latent diffusion model from AI Forever with a large language-model text encoder, supporting both generation and inpainting.
diffusionopen-weights -
Latent Consistency Models (LCM / LCM-LoRA)
2023
Distills diffusion models for 2-4 step generation; the LCM-LoRA variant accelerates existing SD/SDXL checkpoints, enabling near-real-time iteration for art ideation.
distillationlorareal-time -
2023
A DiT-based text-to-image model trained at roughly 10% of SD1.5's cost, with the PixArt-Sigma extension reaching 4K; an efficient open backbone for high-resolution concept art.
diffusion-transformerefficienthigh-resolution -
2023
A distilled SDXL using Adversarial Diffusion Distillation (ADD) for single-step, real-time text-to-image generation, useful for live previews and rapid iteration.
distillationreal-timesdxl -
2023
A latent text-to-image diffusion model with a larger UNet and refiner stage; the open-weights backbone for most game/anime concept-art LoRA ecosystems.
diffusionopen-weights -
2022
The 860M-parameter latent diffusion model that runs on consumer GPUs and anchors the largest ecosystem of anime/game LoRAs, embeddings, and ControlNets.
diffusionopen-weights
Proprietary / Industry Tools¶
-
Recraft Discovered by AI
2023
A professional AI graphic design platform that generates high-quality raster images and fully editable vector art (SVG) from text prompts, tailored for designers and creative studios.
graphic-designproprietaryvector -
Adobe's commercial generative-image API offering text-to-image plus programmatic Photoshop and Lightroom editing operations, marketed as commercially-safe for production use.
adobeapicommercialediting -
A text-to-image model served via AWS Bedrock with image conditioning, color-palette guidance, and built-in editing, convenient for teams already on AWS.
apiawscommercial -
Black Forest Labs' hosted API for the proprietary FLUX 1.1 Pro and FLUX.2 (32B) models, the managed tier above the open FLUX.1/2 weights.
apicommercialflux -
Native image generation and conversational editing through Google's Gemini models, accessible via the Gemini API with text-to-image and image editing.
apicommercialeditinggoogle -
Google's Imagen 4 family on Vertex AI for text-to-image plus editing, inpainting, and outpainting, with enterprise controls and SynthID watermarking.
apicommercialgoogleinpainting -
A commercial text-to-image service noted for reliable in-image text and typography, useful for posters, logos, and title cards.
apicommercialtext-rendering -
A commercial generation platform (text-to-image, image-to-image, image-to-video) with game-asset-oriented models and webhooks, popular in indie game art pipelines.
apicommercialgame-art -
A commercial text-to-image service known for high aesthetic quality, widely used for rapid concept-art ideation across game and anime studios.
commercialconcept-art -
A text-to-image and image-to-image REST API from MiniMax supporting batch generation.
apicommercial -
OpenAI's natively-multimodal image models (gpt-image-1 / 1.5 / mini) for text-to-image and instruction editing, with strong prompt adherence and in-image text.
apicommercialeditingopenai -
A free, open generative-AI platform offering a no-signup text-to-image API, handy for prototyping and hobby projects.
apifree -
A commercial generator producing both raster and true vector (SVG) output via its V4 model, suited to icons, logos, and 2D UI/art assets.
apicommercialvector -
Stability AI's hosted REST API for Stable Diffusion 3.5 and Stable Image models, the managed service alongside the open Stable Diffusion weights.
apicommercial -
xAI's grok-imagine-image text-to-image model exposed via a REST API with an OpenAI-compatible interface.
apicommercialopenai-compatible
