The first fully AI-generated movie just went viral in Cannes.
The film is called Hell Grind.
According to reports, it was made using Higgsfield AI in around 14 days, cost about $500,000 and roughly $400,000 of that went to compute.
For context, a traditional Hollywood feature can take 1-3 years and cost tens of millions to produce.
This one was created by a team of about 15 directors, cinematographers, and editors using AI tools like Higgsfield’s Soul Cinema, Seedance 2.0 and other video models.
The craziest part?
For just the first 25 minutes, CineD reported they generated 16,181 clips to get 253 final shots.
That’s roughly 64 attempts for every shot.
Some prompts were reportedly around 3,000 words each, just to get 15 seconds of usable footage.
So yes, AI is getting faster.
But the real takeaway is not “humans are useless now”
It’s actually the opposite
The people who understand storytelling, directing, editing, taste and tools are about to get insane leverage.
🪄 Introducing daVinci-MagiHuman: The Performance-Level Audio-Video Generative Foundation Model
Proudly open-sourced and jointly developed by SII GAIR Lab & https://t.co/yn4NJpoMrD, it sets a new standard for multimodal AI.
⏳ 1/6
🍌This is #NanoBanana2. Your move.
Hannibal's chilling elegance😈 , Newton's cosmic genius😇 ,
The Scream's raw panic😱 , Dalí's melting eternity ⌚️,
Magritte's Son of Man staring back👁️…
Now it's your turn.
2,000 CREDITS DROP!
Like + RT + Reply "banana"→ 30 lucky creators win big.
Unleash your version on Elser →
https://t.co/zbSSW0zrL7
📅⛵️ Midjourney Office Hours Recap - March, 04 2026
In short: V8 could arrive as early as next week. The model is already launch-ready, but the team is hesitating between a quick release or a few days of polishing. V8 promises tremendous speed, different prompt behavior from V7, and possibly a launch in Turbo mode only.
🇫🇷 Version française en commentaire
🔶 V8 Release Timing
🔹 V8 is “launchable” as soon as next week, but the team is debating whether to ship early with rough edges or delay a few days for polish.
🔹 Current internal question framing:
🔸 What’s acceptable for a “Monday-ish” release vs. a “Friday” release?
🔸 What improvements are realistically achievable with +1 day, +5 days, +2 weeks?
🔹 Likely direction:
🔸 Release something early next week (day not locked).
🔸 Use guides/mods input (and possibly broader community polling) to decide what’s worth delaying for.
🔶 Feature Readiness Tradeoffs
🔹 Key decision: whether to hold the entire release for missing/weak features, or ship and flag/disable them.
🔹 Style references (sref) status:
🔸 Works “okay” in some cases, but can be noticeably worse in others.
🔸 “sref random” seems more reliable than older/specific legacy srefs right now.
🔸 Options being considered:
🔸 Ship with sref limitations (e.g., new srefs only).
🔸 Temporarily disable sref on release if it’s too inconsistent.
🔹 Personalization + aesthetics:
🔸 Multiple aesthetic/personalization “versions” exist internally.
🔸 Considering an “older aesthetics” option to be more compatible with existing moodboards/workflows, even if it slightly limits the new model.
🔹 Core tension: improving prompt adherence and coherence can sometimes reduce “default prettiness,” especially when users write low-quality prompts.
🔶 V8 Model Behavior Changes
🔹 V8 behaves differently than V7:
🔸 Old prompts won’t reliably transfer.
🔸 Some prompts that used to look fine may look bad; some previously weak prompts may become great.
🔹 If released early, expectations management may be explicit:
🔸 The model may change frequently and without notice.
🔸 Workflows may break during rapid iteration.
🔶 Performance, Modes, and Cost
🔹 Speed is a major win:
🔸 V8 is described as extremely fast—faster than V7 Turbo—while aiming to be priced closer to V7 regular.
🔹 Possible launch approach:
🔸 Start with “Turbo-only” (no Relax at launch), then add slower/cheaper options later.
🔸 Pricing may bifurcate later: slower cheaper mode + higher-priced Turbo.
🔹 Resolution / quality controls discussed:
🔸 Native 2048×2048 high-resolution mode exists and looks great, but is ~6× more expensive.
🔸 Q1 and Q4 modes planned:
🔸 Q4 can make runs up to ~24× more expensive when combined with high-res.
🔸 Expectation: Q4 often won’t be useful (may be worse), but could help in niche cases.
🔸 Earlier idea of very low-res (e.g., 512px) “too fast” to fit the current UI, so UI changes are underway.
🔶 Alpha Site and “Danger Zone” Iteration
🔹 Considering shipping V8 via https://t.co/7cWoWvXsFS with a clear “danger zone / alpha” warning.
🔹 Intent:
🔸 Change UI + model frequently.
🔸 Collect examples of “good vs. broken” behavior quickly from users.
🔹 Practical constraint: training vs. inference compute
🔸 Training and generation are competing for the same compute right now.
🔸 Shipping V8 increases inference load and may temporarily reduce training capacity.
🔸 A large new server cluster is expected end of March, which should materially improve capacity and pace.
🔶 Prompting Guidance for V8
🔹 V8 dislikes “cruft prompts” (classic SEO-ish prompt stuffing), e.g.:
🔸 “4K, trending on ArtStation, hyper-realistic, ultra-textured, award-winning…”
🔹 Recommended approach:
🔸 Clean, simple English.
🔸 Long prompts are fine if they’re coherent, not keyword salad.
🔶 Upscalers and Image Pipeline
🔹 Native high-res is strong, but upscalers are being trained now.
🔹 Expectation: an upscaler will arrive later (timeline not committed).
🔶 Recent Shipping: Personalization Updates
🔹 New personalization interface has shipped and is encouraged to try.
🔹 New Niji moodboards + personalization are out (noted as “really good” for Niji users).
🔶 Community Feedback and Release Process
🔹 Team wants better signal on what users actually want from releases:
🔸 Some users want early access even with missing features.
🔸 Others expect day-one stability and no breaking changes.
🔹 Possible action:
🔸 Poll/survey to decide which missing features justify delay (and to set expectations for an “early” release).
🚨 800 FREE CREDITS — 24 HOURS ONLY
We just launched LTX-2.3, our most powerful video model yet. High-res. Fast. Cinematic. Native lip-sync.
Follow + Retweet this post to get 800 credits sent to your DMs.
King Kong vs Tyrannosaurus rex – Ultimate AI Fight (Seedance 2.0)
This is wild.
AI just made a fight Hollywood never dared to make.
Seedance 2.0 is going to change the fate of Hollywood
📅⛵️Midjourney Office Hours weekly Notes - February, 11 2026
In few words: Style Creator is evolving but raises questions about control and UX, SREFs remain fundamental, V8 is advancing quietly, Describe could make a comeback in a few months, and the team is preparing a phase of overall improvement for the site.
🇫🇷 Version française en commentaire
🔶 What’s Shipping Now
🔹 Scrolling Style Creator (recent launch)
🔸UI was updated to be more intuitive.
🔸Team wants heavy user feedback; they plan to keep investing in the scrolling-style workflow and iterate quickly.
🔶 Website Cleanup and Usability
🔹 Focus is shifting to “paper cuts” and general web usability improvements.
🔹 They’re actively collecting the annoying/friction points and expect faster rollouts of fixes.
🔶 V8 Status
🔹 Most effort is on V8, but no details shared (“we don’t talk about V8”).
🔶 Style Creator Future Direction
🔹 Style Creator isn’t being removed soon.
🔸It will likely exist in current form for a while, then be integrated into a broader “Create” flow as a setting.
🔸Goal: creation feels more natural inside the scroll-and-click experience.
🔶 SREFs Are Staying
🔹 SREFs are not going away and aren’t planned to change fundamentally.
🔹 Team views SREFs as a core part of creation and widely loved.
🔶 Key User Pain: Getting Step One Right in Style Creator
🔹 Strong feedback that the current landing/portal approach:
🔸Shows too few results (e.g., ~8–10) and they’re too small to evaluate well.
🔸Can feel like you’re forced into an early seed before you’ve had enough choice.
🔸Makes the whole session feel “locked-in” too early, even if the system is intended to remain steerable.
🔹 User request pattern:
🔸Prefer enter creator immediately, then browse many options (50–100+) to choose a precise starting style.
🔸Want search inside the Style Creator session (not only on the portal) to steer mid-session.
🔸Want to be able to “change direction” (e.g., cartoony → add papercut look) by searching and injecting new style options during scrolling.
🔶 Team Response on “Locked In” Feeling
🔹 Their intent:
🔸Scrolling + selections should keep producing new/refined style codes over time.
🔸You should be able to steer later and still change course.
🔹 What they took away:
🔸Users are reporting it converges too quickly and feels like too little control.
🔸They plan to review algorithm/UX tweaks to make it feel less prematurely narrow.
🔹 UI-specific note:
🔸They agree the seed search results should be larger (visibility issue acknowledged).
🔶 How Style Creator Inputs Actually Affect Results
🔹 Text prompt
🔸Affects the preview images on the right (the prompt context + which pre-render category to display).
🔹 Seeded style/SREF selection
🔸Affects the scrolling grid you refine through.
🔹 Extra clarification:
🔸Prompt words influence which of the pre-rendered example categories gets shown for a style (e.g., portrait vs landscape vs building/etc.).
🔹 Popular request:
🔸Let users switch the preview category for the style-selection grid within the same session (without restarting).
🔶 Refine Button Removal and “Endless Scroll” Feedback
🔹 User request: bring back a Refine Style button so refinement happens when the user triggers it (vs “scroll to trigger”).
🔹 Team stance:
🔸They’re leaning into scroll-based interfaces for now.
🔸They recognize tension between fine-grained control and simplified UX.
🔶 SREF Organization and Recall Problems
🔹 Current state:
🔸Style/SREF organization is acknowledged as not great and a common request, but not top priority.
🔹 “Reverse SREF display” request:
🔸Want to click an SREF code and see a preview/thumbnail (especially when using 5+ SREFs in a prompt).
🔸Team has prototyped this but ran into issues, especially with:
🔸SREF random
🔸numeric codes without a clear canonical preview (“what image should represent it?”).
🔶 Profile Ranking Workflow Tip
🔹 Request: during pair-ranking for a profile, allow selecting an encountered image for another profile.
🔹 Response:
🔸Not supported in current setup.
🔸Workaround: save images you like and build a mood board to capture that “profile personality.”
🔶 Edit Model Surveys
🔹 Two similar surveys were intentionally vague to learn:
🔸What people actually want from editing (character rotation/lighting, object transfer, heavy style transfer, etc.).
🔸How many image references/context users expect.
🔹 These answers affect:
🔸Model architecture choice
🔸dataset/training considerations
🔹 “Character consistency” remains one of the most upvoted long-running community needs.
🔶 “Describe” Feature Update
🔹 There’s a real chance of an updated Describe in ~3–4 months (not a guarantee).
🔹 Likely needs an in-house tuned approach so it matches how Midjourney prompting actually works.
🔹 Priority order implied: after V8 + edit model work, when research bandwidth opens.
🔶 Organization and Power-User Tools (Downloads, Saving Styles)
🔹 Requests heard:
🔸Better organization tools
🔸Improved download system for large exports
🔸Better ways to save/track custom styles
🔹 Team sentiment:
🔸They want to get through the next simplification phase, then add lightweight power-user organization features.
🔶 Competitors and Feature Inspiration
🔹 Team members try competitor tools; they’re not “anti-competition.”
🔹 Perspective shared:
🔸Midjourney still aims for the most beautiful images.
🔸Some competitors excel at specific use cases (example given: meme-making).
🔹 Product view:
🔸They evaluate anything users spend time with, paid or free, to decide what MJ should build.