Learn to create.
Free.

Do this

• Follow the formula. Name the subject, what it's doing, where, in what style, and the shot/light. Missing pieces get filled in randomly.
• Be concrete. "A rusted red bicycle leaning on a brick wall" beats "a nice old bike". Specific nouns and adjectives steer the image; filler words don't.
• Iterate one variable at a time. Change only the lighting, or only the camera angle, then regenerate. If you change five things you won't know which one helped.
• For video, describe motion. Say what moves and how the camera moves — "slow dolly in, subject turns to face us". A still-photo prompt makes a stiff, barely-moving clip.
• Put the important thing first. Earlier words carry more weight, so lead with your true subject, not the background.

Do this

• Think in shots. Storyboard your idea as a handful of 5-second beats, generate each one, then join them — instead of hoping one prompt delivers a full scene.
• Keep motion simple. One clear action per clip stays coherent. Two people doing different things, plus a fast camera move, is where hands and faces fall apart.
• Direct the camera. Name the move — slow pan, dolly in, static locked-off, orbit. "Static" is the safest for clean results.
• Choose the right model. Cheap first: Pixverse v6 (27cr) and Veo 3.1 Lite (54cr) are for finding the shot. Wan 3.0 (54cr) adds native sound and runs to 30 seconds. Step up to Kling v3 Turbo (121cr) or Seedance 2.0 (264cr) once the idea is right, and only then to a 4K lane.
• Draft cheap, finish nice. Settle composition and motion on a cheap engine, then re-run the winner once on a premium one. Ten drafts on Pixverse cost less than one Seedance render.
• Match the shape of your source. On image→video, feed a picture already in the shape you want out. Asking for a tall 9:16 clip from a wide photo forces the model to invent the missing frame, which is where slow renders and failures come from. Some lanes ignore the ratio setting entirely and simply follow your image.

Do this

• Pick aspect ratio for the destination. 9:16 for TikTok/Reels, 16:9 for YouTube, 1:1 for feed posts. Generating in the final shape avoids ugly cropping later.
• Use a reference image. Feed in a picture to carry a style, a color palette, or a character's look into new images — the fastest way to stay consistent across a set.
• Prompt the lighting. golden hour, soft window light, hard rim light, neon — light changes the mood more than almost any other word.
• Name a lens or shot. "85mm portrait, shallow depth of field" or "wide 24mm establishing shot" gives you real photographic control over framing and blur.
• Upscale last. Get the composition you love first, then upscale for resolution. Upscaling a bad composition just gives you a bigger bad image.

Do this

• Save the character once. Create a character asset from a reference image and reuse it, instead of re-describing the person in every prompt and getting a stranger each time.
• Lock the identity, change the scene. Keep the character reference fixed and only change the setting and action — that's how you get the same hero in ten different places.
• Handle wardrobe deliberately. Define outfits as part of the character so clothing stays consistent, or intentionally swap outfits when the story calls for it.
• Use a clean reference. A sharp, well-lit, front-ish face makes a far more reliable character than a blurry side profile — garbage in, inconsistent out.
• Consistency is what makes a series. Audiences follow characters. A recognizable recurring face is what turns one-off clips into content people come back for.

Do this

• Give it a clean subject. One clear object or character, fully in frame, nothing overlapping it. Busy scenes confuse the reconstruction.
• Use a plain background. A flat, uncluttered backdrop helps the model separate subject from surroundings — a cut-out or studio background is ideal.
• Shoot front-ish. A straight-on or slight three-quarter angle gives the model the most to work with. Extreme angles leave hidden sides it has to guess.
• Even lighting helps. Avoid deep shadows that hide geometry; soft, even light reads as true shape rather than as a dark dent.
• Generate the source image first. Make a clean image in the image tool, then send your best frame into 3D — don't feed it a messy screenshot.
• Pick the engine for the job. Tripo H3.1 (48cr) carries real PBR textures, so it is the one for game and product work. Hunyuan3D Rapid (65cr) is the cheap draft. Seed3D 2.0 (77cr) exports USD/USDZ for Apple AR and Blender pipelines. Hunyuan3D Pro (108cr) is the clean-up pass once the shape is right.
• Two engines have a hard limit — know them before you burn a generation. Rapid is text-only (it will not take your image) and Seed3D is image-only (it will not take a prompt). The other two accept either.
• Plan or pack changes the maths. On a 3D plan, one generation is one model whichever engine you choose — so use the expensive engines. On pack credits you pay per engine, so draft on Tripo or Rapid and save Pro for the keeper.

Do this

• Match the anatomy. Human motion onto a human character, four-legged onto four-legged. Driving a human dance onto a creature with a different skeleton produces broken, twisting results.
• Pick a clear reference. One person, full body in frame, steady framing. Crowded or partly-cropped reference video gives the model an unclear motion to copy.
• Keep the moves in range. Simple, readable motion transfers cleanly; extreme acrobatics and fast spins are where limbs detach.
• Use the keep-sound option when you want the reference audio (music, speech) to carry over with the motion instead of a silent clip.
• Start from a strong subject image. A clean, well-posed character image gives the motion something solid to move — a messy subject warps as it animates.

Do this

• Match mood to visuals. Tense scene, tense music; playful scene, bright music. A mismatch (upbeat track over a somber shot) instantly feels wrong to viewers.
• Layer purposefully. A music bed sets the tone, SFX sell specific moments (footsteps, whoosh, impact), and voice carries information. Decide which each clip needs.
• Mind the timing. Line up hits — a beat drop or a sound effect on the exact frame of an action — to make edits feel intentional.
• Keep voice clean and clear. For narration or dialogue, keep the music quieter underneath so the words stay legible.
• Know your plan's limits. Sound generations are metered: Free = none, Creator = 5, Pro = 100, Max = 500, Agency = 1000. Plan your audio passes so you don't burn the allowance on tests.

Do this

• Trim in the editor. Cut the weak first and last frames — AI clips often start or end with a little warp, and a tight trim hides it.
• Extend to lengthen a good clip. If a shot is working but ends too soon, use to continue it rather than regenerating and losing the take you liked.
• Upscale the winner only. Run upscale on your final chosen clip for a crisp, higher-resolution result — not on every draft.
• Upscale vs regenerate. Upscale when the content is right and you just want more resolution. If the content itself is wrong (bad motion, wrong subject), regenerate — upscaling can't fix that.
• Assemble last. Bring your trimmed, extended, upscaled shots into the editor and sequence them into the finished piece.

Do this

• Post to the community feed. Publish your creations so other JOVON users can find, react to, and follow your work — visibility starts with actually posting.
• Stay consistent. A recognizable style, character, or theme makes people remember you. Random one-offs are forgettable; a through-line builds an audience.
• Keep a cadence. Posting regularly beats posting once and vanishing. A steady drip keeps you in front of the feed far more than a single burst.
• Lead with your best. The first frame is the hook — put your strongest, clearest shot up front so people stop scrolling.
• Share the link, not the file. Every post you publish gets its own public page that works for anyone, signed in or not — so you can drop it in a DM, a portfolio or a bio without re-uploading anywhere.

Do this

• Learn the rough rate. With 1 credit ≈ $0.01, a job that costs 40 credits is about 40 cents. Now you can judge whether a re-roll is worth it.
• Start free. Every new account gets 25 images and 3 videos, no card — enough to learn the tools before you decide anything. They are separate allowances, so using images never eats your videos.
• Then a plan or a pack. When the free creations run out, ongoing creating needs a paid plan, a credit pack, or the Diamond Paradox book bundle (which includes 5 videos). Plans run on credits — spend them on something you actually want rather than careless tests.
• Plans vs packs. A plan is a recurring monthly allowance (best if you create regularly); a credit pack is a one-time top-up (best for an occasional burst). Match it to how often you actually create.
• Budget cheap-to-expensive. Draft on the cheap engines — images cost 1–31 credits, and Pixverse v6 is 27 for five seconds of video. Save the 121–454 credit engines for the take you have already framed.
• Add-ons are separate. Extras like 3D are paid add-ons on top of your base plan — factor them in when you plan a project that needs them.

Do this

• Check the prompt against the content rules first. By far the most common failure is not a bug — it is the model refusing. Every image and video model on JOVON is run by a major provider (Google, OpenAI, xAI) and each enforces its own policy. Sexual content, real public figures, graphic violence and copyrighted characters get declined before a single frame is drawn. The job fails instantly, and it will fail every time you retry the same prompt.
• Rewrite rather than re-roll. If a prompt was refused, running it again changes nothing. Change the wording — describe a fictional person instead of a named one, describe the mood instead of the act — and it usually goes straight through.
• Give reference images a fair chance. For character and edit work, use a clear, well-lit, front-facing image where the face is a decent share of the frame. Tiny, dark, heavily filtered or group photos are the second biggest cause of failure.
• Keep references reachable. A reference has to be an image our servers can actually load. If you paste a link from a site that blocks hotlinking, the job dies at the download step. Upload the file instead.
• Let it finish before you leave. Video renders are polled after they are submitted. Closing the tab immediately can leave a job showing as unfinished even though it completed.
• Try a different model. The same prompt can pass on one model and be refused by another — they do not share the same rules. Switching is often faster than rewriting.

What it costs you

• A failed job is not supposed to cost you credits. When a generation fails, the credits reserved for it are returned to your balance automatically.
• Check your balance if something looks wrong. Your balance is in the top bar on every page, and every charge and refund is recorded against your account.
• If a failure did take credits, tell us. Email jevonsand@gmail.com with roughly when it happened. We would rather hand credits back than have you quietly lose them — and we do read every one of these.

The mistake to avoid

Burning ten attempts on a prompt the model was never going to accept. If two identical retries both fail, the prompt is the problem, not the service. Change one thing and try again.

What you get for $9.99

• The book itself — readable on the site, and yours to download once it is attached to your account.
• 5 free Seedance videos — added to your account when the purchase clears, on top of anything you already have.
• 10% off anything on JOVON — a one-time code, emailed to you, with no expiry. Save it for whatever you buy next.

Worth knowing

• The 10% code and the launch discount can't be combined. Stripe accepts one code per order, so use them on two separate purchases — that is why the book code has no expiry.
• Already on a paid plan? The book is $8.99 for you, applied automatically at checkout.
• Sign in before you buy. The book, the videos and the code all attach to your account, so the purchase needs one to attach to.