@ImmaginAI

AI visuals, agents and automations for real business. Let's turn ideas into assets, workflows and working systems. Built by AIDoor.

 Italy
Joined December 2024
  • Tweets1.8K
  • Following556
  • Followers555
“Not complete, Yet we breathe…” Imperfect tools. Unwritten possibilities. What if the end was only the seed? The prototype of a myth still being written. BIAnT is an open lore in the making. Soon open source, yours to shape. Bring your assets, your scripts, your vision. Enter BIAnT. Let’s evolve together.
1
1
29
39,509
1/2 One brief. A creative handoff with receipts. Our Astra kitchen test tracks constraint checks, human corrections and files that actually open. The attached run shows the work, including the rough edges. 🧪
2
39
2/2 Test recipe: public references, one Italian night-scene brief, a shot list, image prompts and a QA checklist. Astra handles planning; Higgsfield handles rendering. Run notes separate the model, tools and human edits.
15
1/2 GPT-6 Astra is rolling out. Our creator test: can it turn a brief into usable work, keep the constraints, and show its receipts? This week, the Digital Kitchen puts the handoff under the spotlight. 👨‍🍳
1
2
75
2/2 OpenAI describes stronger computer use and long-task context in Codex. Those are vendor claims, not our test results. Launch details and rollout limits: openai.com/index/gpt-6-astra…

GPT-6 Astra: A new generation of intelligence

Introducing GPT-6 Astra, our most intelligent and aligned model yet, with state-of-the-art capabilities across computer use, coding, cybersecurity, and science.

openai.com
2
35
Steal this recipe 🍋 A miniature [Italian street] inside a cut-glass [object], rain-polished stone, midnight violet, cyan shop light, one coral window, cinematic macro photography, tactile detail, no text. Swap the place. Keep the light. Show us your tiny world.
1
28
Fresh from the Stove: camera direction just got a spatial pantry. 🎥🍝 Atlas lets a reference image become a world you can move through, with the camera path leading the plate.
1
2
58
World Labs says Atlas takes one or more reference images, lets creators specify camera position and angle, and can generate up to 1 minute of video at 1440p. Atlas is in early access with select partners. worldlabs.ai/blog/atlas

Atlas: A World Model for Spatial Intelligence

Introducing Atlas, our new omni world model for spatial intelligence.

worldlabs.ai
2
34
Chef's Shortcut: Write the mood before the objects. “Electric midnight, lacquered chrome, slow dolly” gives the model a kitchen to work in. Add the subject last, like basil. 🍝
1
28
Fresh from the Stove: Google Pics puts image work where a brief already lives. Make a first pass in Docs or Slides, then polish only the winner. Less tab-hopping, more plating. 🍝
2
33
ImmaginAI by AIDoor retweeted
ImmaginAI by AIDoor
@ImmaginAISep 3
Replying to @BeanieBlossom
Even after dark, does the harbour still glow the color of fresh mussels? 🌌
1
2
4
56
Weekly Recipe: Give one concept three plates: editorial portrait, product still life, motion opener. Keep the same palette, light, and one visual verb. Your feed becomes a menu, not a mystery. 🍅
1
30
The best local model is the one your machine can finish. Start with usable VRAM, context target, quant level and latency budget. Then choose the checkpoint. Drop your GPU and task. We’ll map the fit.
2
69
Abliteration changes behavior. Test the delta across refusals, task accuracy, hallucination and unsafe compliance. A single success clip hides the trade. Which metric belongs in every lab report?
1
59
Hy4 preview helped optimize the system that runs it. Tencent says the open-weight model proposed and tested inference changes that lifted throughput 31.8%. The model is impressive. The recursive engineering loop is the bigger signal. 🔧
Tencent Hy
@TencentHunyuanAug 28
🚀 Hy4 preview is here. 770B, 49B active, 1M context. Built for productivity. Open source frontier. Consistent affordable price. Use it. Tell us what breaks. More on Hy blog：hy.tencent.ai/research/hy4-p… HuggingFace：huggingface.co/tencent/Hy4-p… Github：github.com/Tencent-Hunyuan/H…
1
85
Ox Alpha spent a week running in public, then revealed its name and weights. GLM-5.3-Flash turns stealth testing into an open-model launch loop: real traffic first, MIT weights next. The release playbook may be as important as the model. 🐂
Z.ai
@Zai_orgAug 26
Introducing GLM-5.3-Flash - Leading capabilities at a highly competitive price - Natively multimodal with a 1M-token context window - A 320B-A18B model released under the MIT License - Previously previewed as Ox Alpha, running entirely on Chinese AI chips Blog: z.ai/blog/glm-5.3-flash Available now across all official platforms: Weights: huggingface.co/zai-org/GLM-5… API: docs.z.ai/guides/llm/glm-5.3… Coding Plan: z.ai/subscribe ZCode: zcode.z.ai/en Chat: chat.z.ai AutoClaw: autoclaw.z.ai
1
83
Slides are becoming storyboards. Wan3.0 can turn text, images, audio, video, documents and web pages into clips up to 30 seconds, now via a generally available API. The creative interface just got much wider. What would you feed it first? 🎬
1
58
Video models are learning the language of sequences. LTX-2.5 brings native multi-shot generation, stronger continuity, real-footage editing and open weights into one creative stack. What scene would you direct first?
1
107
A model checkpoint can move closer to native deployment. NVIDIA TensorRT Model Connect turns supported Hugging Face models into end-to-end TensorRT inference in two commands, with bundles that reach C++ APIs. Which model would you connect first?
2
45
NVIDIA is licensing the factory behind the model. Its reported $6B Poolside deal targets the system that built Laguna, a bet that open-weight advantage will come from faster training loops and repeatable iteration. The new moat may be the loop itself. 🏭
1
29
Ternary-weight models meet adaptable edge AI. QVAC Fabric brings LoRA fine-tuning to BitNet b1.58 across heterogeneous GPUs, opening a path toward personalized local models. What would your edge model learn?
1
39
Load more