<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
<title>d4m</title>
<subtitle>A wiki covering generative AI artwork: tools, models, platforms, communities, workflows, and the economics around them.</subtitle>
<link href="https://d4m.io/feed.xml" rel="self"/>
<link href="https://d4m.io/"/>
<id>https://d4m.io/</id>
<updated>2026-11-15T17:50:55.571Z</updated>
<author><name>d4m</name></author>
<entry><title>Midjourney</title><link href="https://d4m.io/midjourney"/><id>https://d4m.io/midjourney</id><updated>2026-11-15T17:50:55.571Z</updated><summary>Midjourney is an image and video generator from a small independent lab in San Francisco. Founder David Holz previously co founded Leap Motion. The open beta of Midjourney began in July 2022. Holz said in August 2022 that the lab was already profitable. Its models are closed, unlike many open diffus</summary></entry>
<entry><title>Training Tool Comparisons</title><link href="https://d4m.io/training-tool-comparisons"/><id>https://d4m.io/training-tool-comparisons</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Training tool comparisons cover open source trainers for adapters and fine tuning. Each comparison weighs kohya ss sd scripts against other trainers on model support, realism and setup effort. Adapter basics are on the LoRA page, and rank settings are in the advanced techniques page. The name kohya </summary></entry>
<entry><title>Multi-LoRA Stacking Workflows</title><link href="https://d4m.io/multi-lora-stacking-workflows"/><id>https://d4m.io/multi-lora-stacking-workflows</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Multi LoRA stacking workflows combine style, identity, and concept adapters within one diffusion generation in ComfyUI. Stable results depend on solo testing, fixed seeds, and small stacks of 2 to 4 modules. Flux adapters apply to the model only, while older checkpoint families also use CLIP strengt</summary></entry>
<entry><title>IP-Adapter production workflows</title><link href="https://d4m.io/ip-adapter-production-workflows"/><id>https://d4m.io/ip-adapter-production-workflows</id><updated>2026-11-15T08:49:25.885Z</updated><summary>IP Adapter and ControlNet were the classic production workflows for steering diffusion. IP Adapter copied a reference image's style or face, while ControlNet locked pose, depth or edges. By late 2026 IP Adapter is legacy, and ControlNet survives as all in one checkpoints for newer open weight models</summary></entry>
<entry><title>Civitai Training Post-Mortems</title><link href="https://d4m.io/civitai-training-post-mortems"/><id>https://d4m.io/civitai-training-post-mortems</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Civitai training post mortems are a distinct genre: creators publish failed runs with full settings, so others can see which change fixed what. Success only tutorials omit that evidence. Each Civitai post mortem pairs a failure with its fix. The settings consensus lives in Civitai community training</summary></entry>
<entry><title>AnimateDiff production adoption gap</title><link href="https://d4m.io/animatediff-production-adoption-gap"/><id>https://d4m.io/animatediff-production-adoption-gap</id><updated>2026-11-15T08:49:25.885Z</updated><summary>AnimateDiff is an open source motion module for animating personalized text to image models. The AnimateDiff production adoption gap describes the contrast between mature tooling, including MotionLoRA checkpoints, and the lack of publicly documented professional use. AI meets traditional art gives c</summary></entry>
<entry><title>ComfyUI Production Infrastructure</title><link href="https://d4m.io/comfyui-production-infrastructure"/><id>https://d4m.io/comfyui-production-infrastructure</id><updated>2026-11-15T08:49:25.885Z</updated><summary>ComfyUI production infrastructure covers serverless deployment, autoscaling, workflow patterns, freelance contracts, and licensing limits. Hosting prices are public, while agency contracts stay opaque. Market data and the production reality check describe this market. RunComfy Serverless turns a clo</summary></entry>
<entry><title>ComfyUI node creators</title><link href="https://d4m.io/comfyui-node-creators"/><id>https://d4m.io/comfyui-node-creators</id><updated>2026-11-15T08:49:25.885Z</updated><summary>ComfyUI node creators maintain the packages that many installations depend on. Their projects cover package management, utility nodes, interface additions, and workflow organization. The companion workflow creators page covers artists who publish reusable graphs. The ComfyUI Manager repository handl</summary></entry>
<entry><title>Platform Migration Patterns</title><link href="https://d4m.io/platform-migration-patterns"/><id>https://d4m.io/platform-migration-patterns</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Platform migration patterns describe how visual artists moved between services after AI training policy changes. One migration wave followed a photo sharing disclosure in May 2024. A second followed Grok image editing on X in December 2025. Current standings appear in Cara vs Bluesky vs Instagram/X.</summary></entry>
<entry><title>Civitai Creator Earnings</title><link href="https://d4m.io/civitai-creator-earnings"/><id>https://d4m.io/civitai-creator-earnings</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Civitai creator earnings come from Buzz, the site currency paid when people generate with, tip, or buy access to a creator's models and LoRAs. Cash payouts run only through a monthly pool for paid members. The old compensation and tips system was retired at the end of August 2026. Bankable rules cha</summary></entry>
<entry><title>Japan</title><link href="https://d4m.io/japan"/><id>https://d4m.io/japan</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Japan regulates AI through a framework law with no fines or other penalties. Domestic computing, semiconductor investment, and research that reduces foreign reliance anchor national strategy. Copyright rules favor AI builders, while anime and manga artists worry about mimicry. Regional context appea</summary></entry>
<entry><title>Image generation frameworks</title><link href="https://d4m.io/image-generation-frameworks"/><id>https://d4m.io/image-generation-frameworks</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Image generation frameworks span local node based tools, enterprise suites, and platform assistants. Local execution runs through ComfyUI Core, while the open source image generation breakthroughs of February 2026 cover the open frontier. Community tool adoption patterns 2025 2026 document workflow </summary></entry>
<entry><title>Automatic1111 / Forge</title><link href="https://d4m.io/automatic1111-forge"/><id>https://d4m.io/automatic1111-forge</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Automatic1111, shortened to A1111, is a browser based WebUI for running open source image models. Forge is lllyasviel's performance focused fork of A1111. Both projects have seen little recent development, and community forks now carry most of the work, while ComfyUI remains active. A1111 offers mod</summary></entry>
<entry><title>Flux</title><link href="https://d4m.io/flux"/><id>https://d4m.io/flux</id><updated>2026-11-15T08:49:25.885Z</updated><summary>Flux is a family of text to image diffusion models from Black Forest Labs. Flux open weight releases became leading alternatives to Stable Diffusion. Flux also anchors a large LoRA training scene, with many community styles, characters, and product adapters built on its open weights. The first gener</summary></entry>
<entry><title>Techniques Impossible Before AI</title><link href="https://d4m.io/techniques-impossible-before-ai"/><id>https://d4m.io/techniques-impossible-before-ai</id><updated>2026-11-15T08:49:25.884Z</updated><summary>Techniques impossible before AI cover tactile texture synthesis, controlled imperfection, and subtle micro motion. Trend coverage in 2026 describes these techniques as generative tool work that manual compositing could not match. Creative Bloq reports artists combining technology with handmade sensi</summary></entry>
<entry><title>Qwen-Image model releases</title><link href="https://d4m.io/qwen-image-model-releases"/><id>https://d4m.io/qwen-image-model-releases</id><updated>2026-11-15T08:49:25.884Z</updated><summary>This page tracks 2026 releases from the Qwen Image line, with the Anima finetune family, SDXL checkpoints, and video tools as related context. Related coverage includes image generation breakthroughs, AI art model reference, and Kohya SS updates. Portrait checkpoints appear in the CivitAI catalog. Q</summary></entry>
<entry><title>Frontier image models</title><link href="https://d4m.io/frontier-image-models"/><id>https://d4m.io/frontier-image-models</id><updated>2026-11-15T08:49:25.884Z</updated><summary>Frontier image models are the systems at the top of blind human preference boards. In November 2026 the leaders are closed models served through an assistant or API. OpenAI holds first place. The models reference hub covers the full landscape, and other striking models sit one tier down. On the Arti</summary></entry>
<entry><title>ComfyUI quantized acceleration</title><link href="https://d4m.io/comfyui-quantized-acceleration"/><id>https://d4m.io/comfyui-quantized-acceleration</id><updated>2026-11-15T08:49:25.884Z</updated><summary>ComfyUI is a node graph tool for local image and video generation. Quantized acceleration, weight streaming, and ROCm support decide which models fit on a given GPU. These features link to Local Generation UIs and to the wider ComfyUI core project. NVIDIA announced a ComfyUI optimization at CES in J</summary></entry>
<entry><title>Open-Source Video Generation Toolkit</title><link href="https://d4m.io/open-source-video-generation-toolkit"/><id>https://d4m.io/open-source-video-generation-toolkit</id><updated>2026-11-15T08:49:25.884Z</updated><summary>The open source video generation toolkit centers on Wan 2.2, specialist models and FramePack. Each tool in the toolkit fills a different role. Wan 2.2 is an open generalist family, and FramePack, from the developer with the handle lllyasviel, covers clips on modest GPUs. Proprietary alternatives app</summary></entry>
<entry><title>Luma AI Dream Machine</title><link href="https://d4m.io/luma-ai-dream-machine"/><id>https://d4m.io/luma-ai-dream-machine</id><updated>2026-11-15T08:49:25.884Z</updated><summary>Luma AI Dream Machine was the original name of Luma's hosted video generation product. Luma's own LLM info page now lists the newest Ray model as the current video model and calls the older product deprecated. This page covers the current Luma lineup, its funding and compute plans, and its assistant</summary></entry>
<entry><title>Optimizers</title><link href="https://d4m.io/optimizers"/><id>https://d4m.io/optimizers</id><updated>2026-11-14T17:53:35.302Z</updated><summary>Optimizers are the rules that turn gradients into weight updates during training. Every LoRA or full fine tune of an image model runs one. The optimizer sets memory use, speed and how sensitive a run is to the learning rate. Plain stochastic gradient descent steps against the gradient with a single </summary></entry>
<entry><title>Diffusion model</title><link href="https://d4m.io/diffusion-model"/><id>https://d4m.io/diffusion-model</id><updated>2026-11-14T17:52:37.244Z</updated><summary>Diffusion models create images by iteratively denoising random noise. During training the model watches images get progressively corrupted with noise and learns to reverse each step. At generation time it starts from pure noise and, guided by a prompt, removes noise over a few dozen steps until an i</summary></entry>
<entry><title>Prompt</title><link href="https://d4m.io/prompt"/><id>https://d4m.io/prompt</id><updated>2026-11-14T17:52:37.244Z</updated><summary>A prompt is the text (or image) input that conditions a generation, and it is the primary creative interface of generative art. The model does not read a prompt as a human would. The text encoder converts it into embeddings, and the encoder's vocabulary and biases limit what the words can do. Encode</summary></entry>
<entry><title>CFG</title><link href="https://d4m.io/cfg"/><id>https://d4m.io/cfg</id><updated>2026-11-14T17:52:37.244Z</updated><summary>Classifier free guidance (CFG) controls how strongly an image follows the prompt. In standard sampling, the model makes two predictions at each denoising step. One is conditioned on the prompt. The other is unconditioned, or conditioned on the negative prompt. The sampler extrapolates past the condi</summary></entry>
<entry><title>img2img</title><link href="https://d4m.io/img2img"/><id>https://d4m.io/img2img</id><updated>2026-11-14T17:52:37.244Z</updated><summary>img2img seeds generation from an existing image instead of pure noise. The source image is encoded into latent space, partially noised, and then denoised under a new prompt. The output inherits the source's composition while the model reinvents its surface. The technique is model agnostic and appear</summary></entry>
<entry><title>Text encoder</title><link href="https://d4m.io/text-encoder"/><id>https://d4m.io/text-encoder</id><updated>2026-11-14T17:52:37.244Z</updated><summary>A text encoder is the language model that turns a prompt into embeddings the image model can condition on. It sits between the text and the diffusion model. Its training data and architecture set the limits of what the conditioning signal can carry. Stable Diffusion 1.5 and 2.0 relied on a single CL</summary></entry>
<entry><title>Checkpoint</title><link href="https://d4m.io/checkpoint"/><id>https://d4m.io/checkpoint</id><updated>2026-11-14T17:52:37.244Z</updated><summary>A checkpoint is a saved snapshot of a model's learned weights at one point in training. Training code writes a checkpoint at intervals so a run can resume after a crash. The final checkpoint, or the best one on a validation set, becomes the released model. The term covers language, image and audio m</summary></entry>
<entry><title>ComfyUI Engine and Ecosystem</title><link href="https://d4m.io/comfyui-engine-and-ecosystem"/><id>https://d4m.io/comfyui-engine-and-ecosystem</id><updated>2026-11-14T17:52:37.241Z</updated><summary>The ComfyUI ecosystem supports open source generative art in 2026. Its node graph joins extensions, model managers, frontends, and containers. Example workflow libraries reduce the learning climb; production teams run the same engine in managed deployments. Daily use in 2026 mixes control with frict</summary></entry>
<entry><title>ComfyUI Community Frustrations</title><link href="https://d4m.io/comfyui-community-frustrations"/><id>https://d4m.io/comfyui-community-frustrations</id><updated>2026-11-14T17:52:37.241Z</updated><summary>ComfyUI community frustrations in 2026 centre on memory management, release changes and custom node upkeep. Many of the most discussed frustrations on the ComfyUI ecosystem's issue tracker concern DynamicVRAM, the memory streaming layer that became the stable default in March. The ComfyUI blog descr</summary></entry>
<entry><title>ComfyUI Extension Adoption Mechanics</title><link href="https://d4m.io/comfyui-extension-adoption-mechanics"/><id>https://d4m.io/comfyui-extension-adoption-mechanics</id><updated>2026-11-14T17:52:37.241Z</updated><summary>ComfyUI Extension Adoption Mechanics describes repository growth, custom node distribution, and community discovery during 2026. It places those mechanisms within the ComfyUI Ecosystem, where GitHub projects and a central manager connect developers with users. ComfyUI grew from 69k GitHub stars to 9</summary></entry>
<entry><title>Breaking Changes &amp; User Complaints</title><link href="https://d4m.io/breaking-changes-user-complaints"/><id>https://d4m.io/breaking-changes-user-complaints</id><updated>2026-11-14T17:52:37.241Z</updated><summary>Breaking changes and user complaints shaped ComfyUI discussion from December 2025 through February 2026. Reports describe failed custom nodes, rising memory use, lost workflows, deleted models, and limited rollback options. The best documented case is a January 2026 API change tracked on GitHub. On </summary></entry>
<entry><title>YouTube creators and education</title><link href="https://d4m.io/youtube-creators-and-education"/><id>https://d4m.io/youtube-creators-and-education</id><updated>2026-11-14T17:52:37.241Z</updated><summary>YouTube first creators and educational platforms provide AI art courses, videos, catalogues, and reusable workflows. Olivio Sarikas, Nerdy Rodent, and the AVeryComfyNerd repository anchor this ecosystem. Advanced training techniques and diagnostics shape its curriculum. Olivio Sarikas had more than </summary></entry>
<entry><title>Open-source maintainer funding</title><link href="https://d4m.io/open-source-maintainer-funding"/><id>https://d4m.io/open-source-maintainer-funding</id><updated>2026-11-14T17:52:37.241Z</updated><summary>Open source maintainer burnout and funding shape the sustainability of the software behind generative art tools. Grants, pledges and endowments now direct money to maintenance. Related coverage appears in AI art tool sustainability and tool adoption patterns. The Kubernetes project announced the ret</summary></entry>
<entry><title>India</title><link href="https://d4m.io/india"/><id>https://d4m.io/india</id><updated>2026-11-14T17:52:37.241Z</updated><summary>India combines traditional motifs, rural craft, augmented reality, and domestic AI infrastructure. The national mission funds local compute and foundation models. Public model launches took place in February 2026, alongside an art fair showing craft based work. The mission runs through seven pillars</summary></entry>
<entry><title>Africa</title><link href="https://d4m.io/africa"/><id>https://d4m.io/africa</id><updated>2026-11-14T17:52:37.241Z</updated><summary>Africa's AI art scene is built around events, regional networks, and production tools such as upscaling. Africa hosted its first AI film event in September 2025, and creators across Africa work with generated imagery alongside traditional art. The first edition launched on September 13, 2025, in the</summary></entry>
<entry><title>LoRA Fine-Tuning Adoption</title><link href="https://d4m.io/lora-fine-tuning-adoption"/><id>https://d4m.io/lora-fine-tuning-adoption</id><updated>2026-11-14T17:52:37.241Z</updated><summary>LoRA fine tuning adoption shows three patterns: a clear winner in FaceID, a rough consensus on dataset size, and a split decision between DreamBooth and LoRA. Related histories appear in technique adoption timelines, when techniques became standard, and Civitai Training Post Mortems. IP Adapter Face</summary></entry>
<entry><title>Stability AI</title><link href="https://d4m.io/stability-ai"/><id>https://d4m.io/stability-ai</id><updated>2026-11-14T17:52:37.241Z</updated><summary>Stability AI is the company behind the open weight Stable Diffusion image models. After a near collapse it rebuilt Stability around music, audio and entertainment tools. Its older image models still anchor much of the LoRA and ComfyUI world. Stability releases ran from the first public weights throu</summary></entry>
<entry><title>Autoencoder</title><link href="https://d4m.io/autoencoder"/><id>https://d4m.io/autoencoder</id><updated>2026-11-14T17:52:37.241Z</updated><summary>An autoencoder is a neural network trained to copy its input through a narrow bottleneck. The encoder compresses data into a short code, and the decoder rebuilds the input from that code. Many diffusion models use an autoencoder as their VAE, though recent work shows the best design is still unsettl</summary></entry>
<entry><title>AI art aesthetics and visual trends</title><link href="https://d4m.io/ai-art-aesthetics-and-visual-trends"/><id>https://d4m.io/ai-art-aesthetics-and-visual-trends</id><updated>2026-11-14T17:52:37.240Z</updated><summary>AI art aesthetics in 2026 favor visible imperfection over a single polished look. Visual trends now shift within months, because Midjourney, Runway, and Black Forest Labs each shipped major updates between June and November. Midjourney made V8.2 its default on July 24. Its Edit Model opened to commu</summary></entry>
<entry><title>AI Infrastructure Production Gap</title><link href="https://d4m.io/ai-infrastructure-production-gap"/><id>https://d4m.io/ai-infrastructure-production-gap</id><updated>2026-11-14T17:52:37.240Z</updated><summary>The AI infrastructure production gap is the distance between AI pilots and systems that run in production. Deloitte research measures the gap for agentic AI. Constraints in compute, energy, cooling, budgets, and staffing are documented in Enterprise AI Infrastructure Gap. A Deloitte study reports th</summary></entry>
</feed>
