AIArtEngine v8.5.2 released
AIArtEngine has released v8.5.2, with a full changelog covering v8.5.1 to v8.5.2. The original text did not disclose specific feature changes or parameter details.
AIArtEngine has released v8.5.2, with a full changelog covering v8.5.1 to v8.5.2. The original text did not disclose specific feature changes or parameter details.
In ComfyUI, the author used four simple Japanese prompts (in front of a Burger King, ordering at the counter, eating a burger at a seat, eating a burger at a chain store) to compare the image outputs of Z-Image-Turbo BF16, Flux.2 Klein 9B, Qwen Image 2.1, Krea-2 Turbo and ChatGPT image 2.5. The prompts were deliberately simplified to observe how each model fills in details on its own when the room for cropping and inference increases.
CreatorOS Studio is open-sourced under Apache-2.0, providing one-click video creation based on ComfyUI on a local GPU, consuming no credits and bypassing platform content filtering.
Why it matters: The author built a local tool due to hosting platform review and credit consumption, demonstrating the feasibility of completing the storyboard-to-final-cut workflow locally with open-source models.
Google Gemma officially showcased the experimental project Embedding Gemma Transcribe, which uses the unified audio-text vector space of EmbeddingGemma 2 to transcribe 1 hour of audio locally in about 13 seconds.
Stable Diffusion Art released a tutorial introducing how to use MiniMax H3 Fun ControlNet Union in ComfyUI to copy body movements from a driving video while generating 24fps video and stereo audio.
Why it matters: The author tested the same driving video with different prompts and gives specific tips on avoiding pitfalls regarding model pairing, resolution and strength parameters.
A user used Wan SCAIL for character replacement on an RTX 4080 Super; results were inconsistent, occasionally fabricating details out of thin air, and generation took a long time. The MiniMax H3 workflow they tried performed even worse: either it completely hallucinated and did not include the person from the source image, or the video was acceptable but retained the original face without completing the replacement. The user therefore questions MiniMax's actual performance and seeks usable workflow or alternative model suggestions.
A user who had stopped using image generation for years started using ComfyUI again and tried Qwen Edit 2.1 and FLUX 2 Krea. Qwen's results were acceptable, but they don't know how to prompt to generate multiple different images of the same character. Their use case is to upscale small images of mythological creatures and pixelated small face images found online into a dataset. Their hardware is four RTX 3090s with 96 GB VRAM in total, 64 GB of RAM and a relatively slow Threadripper CPU; they mainly run LLMs day to day and have limited experience.
Addressing the problem that most I2V generators rely on temporal smoothing and that adding motion turns the image into a smooth plastic look, a creator tried using MiniMax Design connected to After Effects via GPT-6 Astra, completing lighting setup before rendering and then running the video through the built-in H3; in testing, grain and detailed textures were preserved fairly well. The author suggests keeping motion amplitude moderate, as violent pans or fast whip pans look stiff, and notes that the workflow is still being refined.
The ComfyUI team released the LoHa adapter MOSS Image MAX ULTRA hires for MOSS-Image-Base, aimed at improving generation at 4MP and higher resolutions. The adapter was trained on 3.5MP to 10MP mixed anime and photo data and is now available on Hugging Face.
Motion designer and editor Raunak Chhabra integrated the FireCut plugin into Adobe Premiere Pro about a year ago, mainly using its automatic silence removal and animated subtitle generation features, and estimates his editing time has nearly halved. He still publishes one video per week, but redirects the saved time toward B-roll, music selection, and topic planning. He sees AI as merely an efficiency tool, with creative direction still in the editor's hands.
Canva opens its in-house image model Craft to everyone, offering it in Dream Lab and via public API; the company says it scores Elo 1340 on DesignArena, ranking 4th overall and 1st in both the people and landscape categories.
The author released a WIP version of IronLeo78 Minimax Power Producer, merging roughly 42 node stages of the MiniMax H3 reference-to-video pipeline—loading, sampling, decoding, saving, and more—into a single ComfyUI node.
After getting a single-video pipeline of transcription → translation → new audio generation → lip-sync working in ComfyUI, a user scaled up to processing 30-40 clips at once and encountered file mismatches: after one task fails, it is hard to confirm whether `video_17` is still paired with `audio_17`, each item's translation finishes at a different time, and re-running a failed task individually is cumbersome. The user asks how to organize batch processing in a larger ComfyUI workflow—for example, queueing per video, saving each stage in separate folders, or passing filenames through to every node.
AIArtEngine released version v8.5.1, with a full changelog covering v8.5.0 to v8.5.1. The original text did not disclose the specific feature changes or parameter details of this version.
PixVerse's official blog introduces using PixVerse V6 to generate a 16:9, 1080p, 10-second hero shot with native sound, then using each::labs' Video API or visual workflow to complete loudness normalization, reframe cropping, opening copy and logo overlays, and cover output, expanding one approved shot into 24 multi-placement ad assets.
AI Canvas v0.10.15 connects global creative assets, storyboard-to-film, and the editing timeline. Image, video, and audio tasks can retrieve their original results after query interruption or save failure. The storyboard table adds a dedicated "Generate Video" column; finished films can be pushed to the timeline with subtitles and voiceover. GPT Image offers "Generation Quality" and "Generation Background" settings by model, supporting declared transparent background requests. The local database is upgraded from v22 to v23; older versions may not be able to open the upgraded database.
The MindStudio team tested using Claude Code with GoKay's Shader Skills to transcribe AI-generated video clips into real-time runnable GLSL shaders. A simple scene of a rover grabbing alien chips was completed in approximately 45 minutes, and can be driven, interacted with, and have objects picked up in a browser.
The MindStudio team tested using Claude Opus 5.5 connected to Unreal Engine 5.8 via the official Unreal MCP to rebuild AI-generated video into fully procedural 3D scenes, without importing any external models, materials, or textures.
Cozy Clay is a free open-source 3D previsualization tool used to build rough 3D scenes, lock camera positions and blocking before generating AI video, then feeding them as reference images or sequence frames into video models like Kling, Veo, and Sora.
Tagtelier's website freely publishes a collection of prompt illustrations generated with Anima. Each image comes with its prompt and generation settings (model, sampler, CFG, steps, size, seed). Prompts are broken down by categories such as quality, number of people, face, expression, clothing, accessories, pose, background, and composition, allowing users to check and copy items individually or exclude specific words.
The author used Animagine XL 4.0 Opt in ComfyUI to generate an anime character in a 'supine, head on pillow, long hair spread out, looking back at camera' pose, at 1152×896 resolution, Seed 202610171, Steps 28, CFG 5, without using LoRA or ControlNet, relying solely on tag combinations for composition.
The author ran MiniMax H3 video generation on a laptop with an RTX 3060 Laptop (6GB VRAM) and 32GB RAM using ComfyUI, selecting the official int8 quantized model minimax_h3_ref2va_pruned_int8_convrot.safetensors and 8-step Turbo LoRA. A 5-second single-image reference video takes about 5–6 minutes, while a 15-second dual-image reference takes about 12–15 minutes.
Testing a character LoRA trained on Anima-Base v1.0 across seven 2B Anima derivative models in ComfyUI on M1 Max 64GB, with the same prompt and seed 42, only RDBT v2.3 preserved Kei's red bow tie in both full-body and half-body images, and drew two characters as four. Hotaru Blend and CapAnima generated images in about 38 seconds at 8 steps, while Base required about 260 seconds at 30 steps.
Testing Nova Anime AM, CottonAnima, AMIXA, and JANIMA—four Anima 2.9B derivative models—on M1 Max 64GB using ComfyUI, remapping a character LoRA trained on 28-block Anima-Base v1.0 to 40 blocks, with 840 out of 948 keys successfully remapped, and all four models preserving character hairstyles.
AIArtEngine released v8.5.0, with complete changelog covering v8.4.3 to v8.5.0.
Hands-on tests of multiple new models for MiniMax H3 cover Lightx2v 1.3, Dasiwa V4, 10Eros Beta6, Remix 0.8 and Singularity V2. The content comes from Chinese ComfyUI hands-on tests and workflow sharing in the Monster community; the same page also lists T8 ComfyUI integrated package V10, Zhenzhen integrated package V4.6B and several trainer images.
The author has released PixText, a model with fewer than 1B parameters (about 0.95B) for fixing garbled text and Chinese characters in generated images. The model is trained directly in pixel space, requires no text encoder or VAE, and uses MeanFlow to complete text repair with only 4 sampling steps. In ComfyUI, users can select a region and enter the correct text, or first automatically detect text and then adjust the suggested boxes; a comparison node shows the before-and-after effect for each region.
Why it matters: To address the problem of small text in generated images being blurred by VAE compression, the author provides a pixel-space repair model and a ComfyUI box-selection text-editing workflow.
The author uses Claude to write scripts and Comic Creator to draw manga, producing AI manga explainer books for workflow studio and comic creator.
A ComfyUI user is looking for a workflow that can generate several-minute-long NSFW videos; most existing workflows max out at about 10 seconds, or require a prompt every 5 seconds to extend. The user also encounters object permanence issues: after a character in the frame is briefly occluded, details such as clothing are incorrectly restored. The user uses Vast.ai hardware, so compute power is not a constraint.
A creator uses a workflow connecting Claude Code to ComfyUI MCP, mainly relying on Qwen Image 2.1 for drafts, design, atmosphere and story, then moving to H3, and plans to try Seedance 2.5 in projects. He has never used paid models and is unsure about workflows, so he asks the community: what tools and models should be chosen when the budget is sufficient, which is the best image model, and what everyone's workflows are like.
A user asks for help in the ComfyUI community: how to put different people into the same image without losing their respective facial identities, and have them pose and do other actions together. The user has tried krea 2 and flux Klein 9 distilled models with a basic image editing workflow, with hardware of an AMD 6570xt, 12GB VRAM and 32GB system memory, and hopes someone can recommend a feasible workflow.
The author used MOSS-Image to test 20 anime and manga styles, covering 1990s cel animation, Shonen Jump covers, Ken Sugimori watercolor, sakuga and kanji typography, and also included a ComfyUI workflow.
A user is asking for help with a ComfyUI workflow, hoping to use one large prompt to automatically split 45 shots (two people talking in the same scene, multiple camera angles, each clip with different audio) and generate them segment by segment, then import them into DaVinci Resolve for editing. Currently he can only manually prompt and render segment by segment at 3-5 seconds each.
Novae has completed her first full vertical short drama and has published it on platforms including IG, FB, and YouTube, while also using Astra to generate a cinematic book video in about 15 minutes. Cassie is producing a visual novel and has released a new film, and her Kickstarter is progressing smoothly. The two are also trying a Blender-to-Minimax H3 video pipeline to reduce costs, and will share it in a livestream at 10 a.m. Pacific Time on Friday.
Kijai's MOSS-Image Edit LoRA support has been merged into the latest ComfyUI Portable, giving MOSS-Image Edit model capabilities and ControlNet-like basic functions. Supporting releases also include the CharRef v2 character reference node, a 9.4Gb Dirty MOSS realistic fine-tune, an animation style LoRA, and a library of 500 usable prompts with example images.
To address the problem that ComfyUI users cannot remember the effects of lighting prompts, the author created a custom node where clicking lets you select lighting expression terms such as backlight, spotlight, God Rays, particles, low-key, and specular reflection, and hovering the mouse also displays text explanations. In the example, after overlaying these terms on an armored character, the image changes from flat to a heavy, solemn atmosphere. The author also notes that newer-generation models such as Krea2 already have fairly high inherent quality, but this node can be used to precisely hit the desired lighting effect.
A ComfyUI beginner is seeking workflow advice, hoping to use Trellis to generate a 3D model from a 2D photo of a person to make a statue of a friend, with results benchmarked against Bambu Lab's "Make My Statue" studio on MakerWorld. His current results with Trellis are far below Bambu's, but he does not want to pay for every attempt and hopes to do it locally.
A developer is building LoRA Studio, a local browser interface for dataset preparation, caption editing and managing LoRA training via AI Toolkit, currently focused on training Krea 2 on 12 GB VRAM GPUs, with 8 GB as a later goal.
Haishen Hyacinth shared a video production workflow combining Claude Opus 5.5 and Seedance 2.5: first use Opus 5.5 to create dynamic storyboards, then produce final-effect keyframes, then use the keyframes as references and hand them to Seedance 2.5 to generate video.
The open-source creation tool released v1.6.9.2; after closing short drama production, the project entry in the sidebar and command palette is now hidden accordingly. This version fixes the proxy address pointer parameter in the channel migration test, restores compilation of the Go test package, and has no database schema changes.